物聯網： 應用與挑戰

黃彥男

中研院資創中心主任

經歷 – 現在

現職

中研院資創中心特聘研究員及主任

中央研究院資安及智慧優網專題中心執行長

亞洲物聯網聯盟理事長

臺灣物聯網產業技術協會， 車聯網委員會召集人

IEEE Fellow

2

經歷 – 過去

AT&T Bell Labs (1126, Software Research Lab): MTS, DMTS

AT&T Labs: Director, Executive Director

PreCache: VP of Engineering

財團法人資訊工業策進會（III）: 副執行長

威達雲端電訊股份有限公司 (VeeTime) : 總經理

行政院科技會報辦公室: 副執行秘書

行政院數位匯流辦公室: 執行秘書

國家資訊通信發展推動(NICI) 小組: 副執行秘書

資策會執行董事, 電信技術中心董事，電腦技能基金會董事，數位聯合電信股份有限公司董事

3

Outline

Introduction

IoT Challenges

IoT Security and Privacy Issues

Industry 4.0

An Example: Using IoT and AI for Preventive

Maintenance

Industry Platforms for Industry 4.0

生活形態的改變

電力綫

電話綫

網路綫

物聯網+人工智慧 -》 智慧生活， 製造

5

物聯網的歷史

1999： Kevin Ashton （MIT Auto ID Center）提出IoT概念

2003 SUN article: Toward a Global“Internet of Things”

2005年11月17日：在WSIS會議上，國際電信聯盟（ITU）發佈《ITU互聯網報告2005：物聯網》

2009年1月23日：美國 Obama 說：物聯網技術美國在21世紀保持和奪回競爭優勢的方式

2009年8月：中國温家寶總理在无锡提出物聯網感知中國

2009年9月：Internet of Things – An action plan for 歐盟行動計畫

2012年12月： Cisco 進一步提出IoE（Internet of Everything）的概念，亦即擴大M2M基礎到人、流程、資料上，並增加新型態智慧型裝置、智慧分析及預測等，將資訊轉為業務優化或有效決策。

6

7

物聯網商機誘人

The World in 202029.5 Billion - Connected “Things”

$1.7 Trillion – Market Opportunity

Asia / Pacific 2020 :IoT’s frontline

8.6 Billion - Connected “Things”

$678 Billion – Market Opportunity

9

NII Program

e-government Program

ICT Policy Roadmap

National Information & Communications Initiative

(2002-2006)

National Information & Communications Initiative

(2007-2011)

‘98 ‘99 ‘00 ‘01 ‘02 ‘03 ‘04 ‘05 ‘06 ‘07 ’08 ‘09 ‘10 ‘11 ‘12

1. e-government2. Industrye-development3. ICT infrastructure

e-Taiwan

M-Taiwan

Digital Content Program

u-Taiwan

1. Mobilecommunicationsconstruction2. Deepen ICTapplications

upgrade

Industrial Automation & Electronic Business Program

First time that government public construction fees were obtained for non-traditional construction projects, to stimulate and accelerate national information and communication application development

Cloud Computing

Initiative

Digital Convergence

Initiative

10

SMART

HOMEEMERGENCY

CARE

ApplicationsID for

every object

FOOD CHAINTRACEABILITY

Converged networks

u-Taiwan

ITS &

TELEMATICS

GOV

SERVICES

Civil Servants

The public

11

Constructing a trusty, convenient, cultural and

healthy ubiquitous network society

Promoting the user-oriented and demand-oriented ICT applications, and to meet the Green IT

Promoting NGN and network convergence (wireless, sensor network, fixed-line, and broadcast integration), and objects connected to network

W

I

S

D

Intelligent Taiwan

O

M

Utilizing the ICT to integrate and innovate government services, and also encouraging citizens’ participation

Developing affordable ICT devices and applicationsfor disadvantaged groups, and to ensure equal e-opportunities

Establishing a solid and sound business environment, and promoting esthetics into citizens’ daily lives.

WISDOM

Enhancing the reading and IT capability, Encouraging the lifelong learning concept, and cultivating next generation manpower

Cultural & Creative Industry

Wireless & Broadband Convergence

Superior e–government

Demand-driven Applications

Equal Digital Opportunity

Manpower Cultivation

5+N 創新研發計畫

連結

未來

連結全球

連結在地

生技醫藥亞洲矽谷智慧機械綠能科技國防產業

新農業循環經濟數位國家文化科技半導體

12

物聯網架構

IoT is the trend

Forrester forecasts by 2020 that things connection to people ratio is 30 to 1, creating the next trillion dollar industry

According to III report there are more than 1000 urban IoT project, and a growth rate more than 20%

14

The IoT age

15

IoT＝Sensing＋Networking＋Intelligence

Source: III

IBM + smarter planet = President Obama’s

National Urban Policy

Sensing China: the 12th 5 year plan focuses

on IoT

u-Japan focuses on IoT infrastructure

i-Japan 2015 focuses e-Gov, e-Healthcare

EU’s future internet project focusing on smart

grid and intelligent transport

IoT Roadmap of Taiwan

From RFID technology-driven applications, to holistic User-

centric i236 projects

RFID e-transaction > 18M cards

16

RFID

e-transaction, e-tagsIntelligent City

Home, Energy,

Healthcare,

Transportation etc.Wireless city

Monitoring

Ap

plic

atio

nD

evic

e

Value Proposition

Environment

Sensing and Control

Sensor and Devices

I236

Smart Town,

Smart Park

Biotech

Tourism

Green

Healthcare

Agriculture

Culture

RFID tag/Reader,

Camera, GPS

High performance

and cost effecting

components

Current State of Taiwan IoT

Device Design Competitiveness with strong roots in IC design, Sensors,

embedded systems, WSN to provide scalable device products.

Information Service Foundation in Smart Grid, Energy Management, Smart

buildings, Smart Homes with business experience accumulation particular in

the e-transaction project.

17

Up stream Midstream Lower -reaches

Chips Modules Machines SI Services

IC, Chip desin(Gateway、Meter、Intelligent

device/server)

Energy management

Data management

Taiwan’s IoT Challenge

18

Sensors & Actuators

Sensor Network

Sensor Gateways

Wide Area Network

Application Gateway

Service platform

Application

Inte

gra

te S

erv

ice

Op

era

tion

To

ols

Qu

ality

an

d S

ecu

rity

Data

to K

no

wle

dg

e

Sensing

Network

Application

Service

deployment

and operation

Data analytics

and domain

experts

SI and

application

provider ?

Business

Innovation

?

Source: III

Taiwan’s IoT Entry Point Strategy

19

Sensin

gN

etw

ork

Applic

atio

n

Application

Service platform

Application Gateway

Wide Area Network

Sensor Gateways

Sensor Network

Sensors & Actuators

Identify key areas for intelligent services

Taiwan’s own Analytics Engine

Lack of scalable chip, thus aim towards higher

value-add devices and gateways

Develop diverse broadband technology

Info

rmatio

n S

ecurity

Pro

ducts

Taiwan IoT directions

Disaster Prevention

Logistics

Energy

Transportation

Healthcare

Living

20

Inte

lligen

t City

Info

rmatio

n S

ecu

rity

IOT 的挑戰

21

IoT 產業特性

缺乏標準，百家爭鳴，互不相容

垂直的整合度高，單一應用很難跨到另一個領域

無主導的大廠

技術仍在演進中

商業模式不明

22

Smart Home 平臺：百家爭鳴

Apple : HomeKit

AMD: Ambidextrous Computing

ARM： mBED

Broadcom: WiCED

Google/Nest: Thread

Intel: Edison, Curie

…..

23

Air Interface Protocols

RFID and NFC

Li-Fi

QR codes.

Bluetooth low energy

6LowPAN

ZigBee

Z-Wave

Thread

LTE-Advanced: NB-IoT, LTE-M,…

WiFi-Direct

HaLow

HomePlug

MoCA

Ethernet

SigFox

LoRa

Neul

IoT 資訊安全的因素

連結的物品多，傳播的訊息量大且複雜， 可能包含個人、商業及政府的隱私機密

設備運算能力有限， 或電力考量

無線訊號在空氣中傳播，容易遭受外部攻擊與干擾

無人監控，設備容易被破壞，盜取或冒名使用

感測器、連網設備未能受到企業防火牆保護

IoT 作業系統沒有自動安全更新的能力

Intranet 物聯網造成資料外洩機率增加

太多不同的通訊協定

設備太多太複雜，很難監控及管理

25

IoT Top 10 Security Infrastructure Issues

Insecure Web Interface

Insufficient Authentication/Authorization

Lack of Transport Encryption

Privacy Concerns

Insecure Software/Firmware

Insecure Network Services

Insecure Cloud Interface

Insecure Mobile Interface

Insufficient Security Configurability

Poor Physical Security

26Source: OWASP2014

IoT 資安威脅的例子

自動車劫持

行車綁架

在特定的街景地形，無人機莫名自動墜機砸傷甚至砸死人

劫持廠房工業用 IoT 裝置：去年夏天，美國研究人員對發力發電公司進行資安演習，以物理入侵方式（破壞門鎖）輕鬆進入該公司旗下一風力發電廠的網路機房，以小小一台Raspberry PI 電腦，關掉五個風力發電廠的所有風力發電機

電力輸配異常

讓特定有慢性病的重要人物失能甚至遇害— 駭入印表機而印出錯誤的醫療指示與病歷，使得一個人被醫死在手術台上

27

物聯網設備是發動DDoS攻擊的完美平臺

非常大量的裝置：IDC預估，在2020年，全球IoT裝置數量將達到300億個

常保可連結性：24x7 連網

具有不錯的運算能力： Raspberry Pi 2 -單顆4核心時脈900MHz的ARM Cortex-A7處理器，以及1GB記憶體

安全不設防的設計：幾乎都沒有內載、安裝安全軟體，怕影響效能

嵌入式系統:設備主要採用Linux或Windows嵌入式系統設計

Example: Mirai, 針對執行Linux的系統

28

SCADA （supervisory control and data acquisition）工控系統

Stuxnet：利用 Simens WinCC/PCS7系统的4個 漏洞， 2010年國際頭條新聞。其專門針對鈾濃縮設施，使離心機失控從而造成納坦茲的工廠遭受物理破壞，成功使正在用來濃縮鈾的1000台離心機癱瘓

Havex： 感染SCADA系統的工業控制系統，這種惡意代碼可能通過使用一個按鍵就能夠使水電大壩停運、核電站過載，甚至關閉一個國家的電網。7/16， 駭客操控工控系統，攻擊歐洲能源公司，偷走汽油及燃煤

IoT Privacy

隱私

32

可能的問題

知道你何時不在

知道你的起居活動

33

34

35

36

37

To release data to third party or even open the

data to the public, the data owner must first

protect sensitive data attributes well to

eliminate the possible of privacy leak.

Data transmission: could be intercepted and

decoded

Both needs to be addressed……….

Privacy Concern for collecting data

38

Collected data from IoT devices is of great

value when properly analyzed.

However, the data owner often doesn’t have

such capability to analyze data, as data

science is an highly sophisticated area.

There’re different type of technology that can

enable data analysis without giving raw data to

the data analyst.

=> De-Identification

Ensuring privacy when releasing data to a

third party or the general public

39

IN 2010, NetFlix included 100 million movie ratings,

along with the date of the rating, a unique ID number

for the subscriber, and the movie info.

Privacy Breach Events (1)

40

“Sparsity” of data: In Netflix data, not two records are

similar more than 50%.

If the profile can be matched up to 50% similarity to a

profile in IMDB , then the adversary knows with good

chance the true identity of the profile.

Found that if you knew a few movies a Netflix subscriber had

rented in a given time period, you could reverse-engineer the

data and find out the rest of their viewing history.

A. Narayanan and V. Shmatikov, “Robust de-anonymization of large sparse datasets (how to

break anonymity of the netflix prize dataset),” in Proc. 29th IEEE Symposium on Security and

Privacy, 2008.

41

De-anonymize Netflix data

The state of Massachusetts distributed a research dataset containing de-identified insurance reimbursement records of Massachusetts state employees that had been hospitalized. To protect the employees’ privacy, their names were stripped from the dataset, but the employees’ date of birth, zip code, and sex was preserved to allow for statistical analysis.

Sweeney was able to re-identify the governor’s records by searching for the “de-identified” records that matched the Governor’s date of birth, zip code, and sex. She learned this information from the Cambridge voter registration list, which she purchased for $20. Sweeney then generalized her findings, arguing that up to 87% of the U.S. population could be uniquely identified by their 5-digit ZIP code, date of birth, and sex based on the 1990 census.

Privacy Breach Events (2)

42

Encryption

How to analyze and compute on encrypted data?

Anonymization

Re-identification is possible

Access mediation/control

With multiple queries, re-identification is possible

Adding noise

Permutation

Differential Privacy

Data De-Identification is difficult

43

Name Has AIDS

Tom 1

John 0

Eric 1

Ross 0

Steve 1

44

Differential Privacy: Basic Concept

f(i) : the partial sum of the first i rows

f(5) – f(4) : reveals Steve has AIDS

Ensure that the removal or addition of any

record in the database does not change the

outcome of any analysis by much.

45

Differential Privacy

Adding random noise to query

f(X) + Lap (∆f/Ɛ)

Ɛ: differential budget

∆f: sensitivity of query function f

IoT devices collect information from all kinds of

sensors and send them through Wifi

connection to remote servers, so it is always

possible that someone try to intercept those

information and decode them.

Differential privacy can also be used to transfer

data securely.

Differential privacy as an option to transfer

IoT data securely

46

47

Raptor Algorithm

Set f=0.5

Do a survey, “Do you have AIDS?”

toss a coin

If Head, always say yes

If Tail, tell the truth

The true percentage of people who participate in the

survey who has is 2(Y-0.5)

48

Example

物聯網應用： 智慧製造

製造業面臨的問題及趨勢

生育率降低、人口老化，勞動人口不足

工資上升快速，成本增加

產品生命周期變短

物聯網、3D列印，大數據等新科技催生智慧製造

歐、美、日強權重新關注在地製造業

環保意識擡頭

個性化（ personalization）需求增加，製造業特性由大量製造（Mass production ）轉向大量客製（Mass Customization）

需數位化，虛擬化，智慧化及自動化來提升效能

50

工業生產的關鍵問題

技術及操作人員

成本

生產的複雜度

彈性製造

預測維護

品質檢測

能源與環境

跨領域合作

52

53

工業4.0

工業4.0也不是單純的工業化+資訊化。利用大資料來分析使用者，瞭解產品背後看不到的規律才是重點。

工業4.0包括製造業6M系統及大數據6C系統

6M in Industry 4.0

建模（Model）、

測量（Measurement）、

工藝（Method）、

設備（Machine）、

材料（Material）、

維護（Maintenance）

6Cs in Industry 4.0

連接 - Connection (sensor and networks)

雲 - Cloud (computing and data on demand)

網路 - Cyber

內容 - Content/context (correlation)

社群 - Community (sharing & collaboration)

客制化 - Customization (personalization and value)

小量多樣

Industry 4.0 Impacts

Services and business models

Reliability and continuous productivity

IT security

Machine safety

Product lifecycles

Industry value chain

Workers' education and skills

Socio-economic factors

Global competition

德國電梯製造業者的故事

德國電梯製造商的客戶，接到中國廠商的洽詢，他們在電梯外部及電機上附加感測器，透過電機和電梯運作的監控訊息判斷電梯是不是即將出問題。在電梯還沒有故障之前，就會趕來提供維修服務，客戶不必等電梯壞了才打電話給電梯公司，然後等上好幾個小時才會有人來修。這家中國來的小公司完全不用經過電梯製造公司，也完全不需要電梯公司掌握的數據和客戶資訊，他自己也不生產電梯，只做服務。換言之，電梯製造業者賣掉電梯之後，就有可能完全被晾在一邊了。

資料來源: 工業4.0 – 58 秒的競爭

An Example: Preventive Maintenance

計畫期程• 自106/05/01至106/12/31止(8個月)。

核定經費• 新台幣約1,000,000元整。

核心目標• 提出有效的預防性維護技術，

以確保工具母機於正常狀態下工作。研究對象

• 工具母機關鍵零部件(直結式主軸)。資料類型

• 三軸振動訊號、聲音、溫度訊號等。解決方式

• 以統計與機器學習等方法執行研究。

智慧型工具機主軸狀態監測與異常診斷系統技術研發計畫

永進自行研製之直結式主軸

永進自行研製之工具機

技術概況

ITRI

實驗架構

實驗情境

出差規劃1~2次/月

轉速條件9000/11250rpm

資料擷取10k取樣,持續4秒

量測情境主軸空跑時進行量測

OK/NG定義客戶反應

其他

強化現行策略量化數值,羅吉斯回歸...

挖掘關鍵指標預測機台壽命

挖掘關鍵指標預測模型輕量化

預期目標

實驗架構

數據類型

2016.08.29

-

-

機號0747-2筆機號0760-3筆

-

機號0968-3筆

共8筆資料

2017.05.24

機號0739-20筆機號0746-20筆機號0747-20筆機號0760-20筆機號0776-20筆

-

共100筆資料

2017.07.26

機號0739-60筆機號0746-40筆機號0747-70筆機號0760-60筆

-

-

共230筆資料

2016.06.08

機號正常-4筆機號注意-6筆機號危險-4筆

共14筆資料

實驗架構

振動(單/三軸)與聲音

訊號蒐集 資料探勘 健康評估

電壓電流(負載電壓)

溫度(軸承,馬達,室溫)

機台資訊(扭距,實際主軸速度)

加工條件

其他(運轉時間)

時域、頻域、時頻域

重力加速度、加速度、速度、位移

統計分析

頻譜特徵

軸心軌跡

其他

羅吉斯回歸

二維分布

模糊推論

高斯混合模型

神經網路

其他

資料檢視

安裝位置:下方側方

案例一

• 步驟一、旋轉軸(X)振動• 步驟二、統計方法• 步驟三、二維特徵分布• 步驟四、異常趨勢觀察

(a) (b) (c)

如標準偏差、均方根值、峰值、波峰因數、峰度、偏度…等

峰度

標準偏差

7月振動強度相較於5月時

• 9000rpm振動減弱了許多

• 11250rpm振動卻暴增

案例二

• 步驟一、徑向旋轉軸(X,Y)振動• 步驟二、軌跡分布• 步驟三、量化• 步驟四、主要分布區域• 步驟五、異常趨勢觀察

(a) (b) (c)

X、Y、Z軸的原始振動訊號

(a) (b) (c)

X、Y、Z軸的振動分布

(a) (c)(b)

X、Y、Z軸的量化表現

案例二

三種狀態的分布區域(1~36)

三種狀態的主要分布區域(14.15.16.20.21.22.28)

案例二

運用模糊理論(特徵分布與主特徵分析)評價機台運轉狀態(正常、注意、危險)。

輸入 推論/學習 決策/預測

其他案例一

運用統計方法(直方圖與關連性分析)評價機台運轉狀態(正常、注意、危險)。

輸入 推論/學習 決策/預測

其他案例二

運用神經網路(統計方法與特徵指標)評價機台運轉狀態(正常、注意、危險)。

輸入 推論/學習 決策/預測

聲音訊號分析工具

20 40 60 80 100 120

50

100

150

200

250

Short Time Fourier

Transform (STFT)

聲音訊號

聲音頻譜能量

聲音頻譜相位

原始聲音訊號，可以看出音量、但是很難分析時頻

較容易看出時頻特性

通常不使用

聲音訊號通常為多個弦波組成，因此原始聲音訊號較不容易分析。最常用的做法是利用時頻特性，分析出聲音時間以及頻率相對特性。時頻分析後通常會產生出相位圖譜以及能量圖譜，我們針對能量圖譜進行分析。

轉速：2000rpm 轉速：11250rpm

轉速慢(左方圖)：聲音較大、無規律時頻特性轉速快(右方圖)：聲音較小、有規律時頻特性

聲音訊號

聲音訊號時頻分析圖

聲音訊號

聲音訊號時頻分析圖

軸承轉速與聲音訊號關聯性之分析

進展：

由振動訊號得知，感測器安裝位置與資料蒐集的狀況

由統計分析得知，轉速與振動特性的關聯

持續進行數據分析，如頻譜、軸軌跡、其他人工智慧方法

擴大訊號源種類，如聲音、溫度、機台資訊、加工條件等

針對送修品進行故障檢驗與資料蒐集

目標：

就現有策略(振動)進行持續性的測試與強化，以足夠且有意義的數據來建構初步的主軸馬達預測維護機制。

在更多訊號源中找尋因果及關連性，建構最接近真實的趨勢預測技術。

進展與目標

工業4.0 Platforms

爲何需要發展發展工業4.0 的 solutions/platforms

Connect various IoT devices

Application developers can access open interfaces and use it for their

own services and analyses –

Online monitoring of globally distributed machine tools, industrial

robots, or industrial equipment such as compressors and pumps.

Customers are also able to create digital models of their plants with real

data from the production process. This allows them to synchronize the

model and the plant, enabling them to carry out simulations and optimize

business processes.

Users will also be able to develop their own web services which can also

serve as a basis for digital services

predictive maintenance

energy data management

resource optimization

Quality prediction

76

換言之

A platform to connect many IoT devices to your

machines so that it can harness big data from billions

of intelligent devices, enabling you to uncover

transformational insights across your entire business.

工業4.0 Platforms/Solutions 的發展方式

工具機公司自己發展工業4.0軟體平臺

工具機 controller company 發展工業4.0平臺

成立軟體公司， 利用雲端架構， 發展工業4.0平臺

自己發展工業4.0軟體平臺

DMG MORI 所發展的 CELOS，

LOKUMA 的 Okuma Smart Factory ，

MAZAK的 iSmart Factory，

FANUC 的 FIELD

Lokuma – Smart Factory

Mazak iSmart factory

MT Connect standard communication protocol

FAUNC FIELD

Controller company 發展工業4.0平臺

Siemens MindSphere

Siemens MindSphere

投資成立軟體公司， 利用雲端架構， 發展工業4.0

平臺

DMG MORI 投資一家公司， ADAMOS

Concluding Remarks

AI + IoT will change how we live and how we produce.

IoT has a lot of challenges

AI + IoT for smart manufacture -> Industry 4.0

IoT Security and Privacy is a big problem to be solved.