using less material with better planning and defect ... 2018... · using less material with better...
TRANSCRIPT
Using Less Material with better planning and
defect detection in Telecommunication and
Manufacturing.
Le Hong Viet
CTO, FPT Corporation
FPT Overview
22
Highest International Standards
SEI CMMI Level 5 | ISO 27001:2013 (BS 7799 -2:2002) | ISO/ TS 16949 |
ISO 9001:2008 | ISO 20000-1:2011
We are where our clients are: 33 global offices
Key partners of Major Technology Providers
#1 in IT
Services/ITO
& Systems Integration in
Vietnam
#1 Entertainment
Portal in
Southeast Asia #1
Private
University
in Vietnam22,000
Students
FPT - The leading ICT
Corporation in Vietnam
Founded: 1988
Public
listed:
Dec 2006 on VNSE
Revenue: USD 1.96 billion (2017)
Staff: 32,092 (Technology
Engineers and
Experts: 15,449)
Developme-
nt Centers:
Hanoi, Hochiminh City,
Danang, Can Tho
R&D investment:
5% profit before tax
Clients: 250+ clients with 40+
Fortune 500 s
© Copyright by FPT 2018
FPT – Leading ICT Company in Vietnam
3
1 | TECHNOLOGY SECTOR 2 | TELECOMMUNICATIONS SECTOR
3 | DISTRIBUTION AND RETAIL SECTOR 4 | EDUCATION SECTOR
Software Development
System Integration & IT Services
#1 in Revenue and Workforce;
#1 in IT services provision;
High level partner of Microsoft, AWS,
SAP, etc
TOP 100 IAOP’s Global Outsourcing;
~12,000 Technology experts & engineers;
#1 in Technology product distribution
1,500+ agents in Vietnam
Distributor for 30+ large technology
partners 450 retail stores in 63/63 cities
and provinces
Telecommunications
#2 in Fixed broadband internet access services;
North-South axis, 8,400km long;
280Gbps international bandwidth;
04 Data centers;
Digital content
34.5 mil. pageviews/day for VnExpress
Leading e-commerce marketplace -Sendo.vn
Top University in IT Education; 17,000
students at all levels (FPT University, FPT
Polytechnic…); 98% of graduates having
jobs within 06 months;
Introduction
A printing company in Japan is wasting $30M a year due to defects (with the rate of 1%) in their printing products and wish to
use Big Data Analytics to reduce this number.
6© Copyright by FPT 2018
Source of Defects
(1) Design and Retouch
(2)Review and Approve
(3)Controlled Printing
(4)Maintenance
- Human Error when designing printing
products.
- Unattended Review, overload of review
tasks
- Machine or Material Defects.
- Control and Monitoring
- Defect parts- Down time
7© Copyright by FPT 2018
(1)
(2)
(3)
(4)
Solution
8© Copyright by FPT 2018
STREAM 1
Human Error
Reduction
STREAM 2
Monitoring &
Alert
• Defect detection sensors
• PLC data integration
• Just-in-time Alert
• Abnormally Detection
• Collect data design
office: (temperature,
humidity)
• Designer behaviors
(mouse trajectory, heart
beat…)
• Supervised Learning via
Review Data
STREAM 3
Predictive
Maintenance
• Usage Sensor
installation
• Predictive Maintenance
Analytic Platforms
Effects / Findings
• System is still in evaluation, however, few important finding has been concluded.
• Random Check is replaced by prioritized review.• Suspected products will be prioritized for review
• Alerts make reviewer be more focused on faulty products.
• Recommended break for Designers, regain focuses.
• Working condition improved.
• Just-in-time alert helps to reduce mass defects.
9© Copyright by FPT 2018
Background
• FPT Telecom owns a largest FTTH network in Vietnam, reaching to multi million households across 63 provinces.
• Wiring the network to households consume a lot of effort and resources.
• Unoptimized Routing and handling causes inefficiency in using fiber optic cables.
11© Copyright by FPT 2018
Process
• Process 1: Running L1 cables, connecting L1 and L2.
• Workers manually draw the cable paths on a map.• Cable are provided with length 115%*L for the
workers.• They then do the laying, with the provided cables,
roughly according to the drawing of step 1.
• Process 2: Running L2 cables to households.• Workers are provided with almost unlimited cable
length.• Workers manually lay the cable from L2 nodes to
customer nodes, actual length is recorded.• The cable length used is compared to distance on
the map, between customer nodes and L2 node, to check if there is excessive spent of cable.
• Randomly, some customer nodes are select for post examination, by independent teams of examiners, to check if the cable running is reasonable.
12© Copyright by FPT 2018
Node L1
Node L2
Customer Customer
Node L2
Customer
Problem & Solution
• Predict Future Consumptions:• Problem: Cost for wiring is high => predict consumption is crucial for
the initial wiring.
• Solution:• Inputs:
• Using previous customer data on Coper Network.
• Using Demographic Survey from Government.
• On the field information about: residential characteristics, income estimation, …
• Solution: Machine Learning models to predict consumptions
• Capacity of network
• Network growth rate.
• Result:
• 3% improvement over experts design.
15© Copyright by FPT 2018
16© Copyright by FPT 2018
Prediction of consumption in 1 year
Pink nodes: need
upgrade to match
future demand
Blue nodes: can
survive the demand
in 1 year
Problem & Solution
• Routing Optimization and Anomaly Detection• Problem:
• Inefficiency in route planning especially in choosing POP location (L2 location)
• Terrain dependence (hanging on electrical popes or underground).
• Tiny roads (not on map).
• Solution:• Workers, besides main work, need to collect GPS, maps, pictures on routes and
meta data (e.g.: Road on map, underground, hanging high….)
• Machine learning algorithms are used to detect abnormal wiring.
• History of wiring (from multiple teams) are used to improve planning.
• Result:• Expecting 5-7% of improvement.
17© Copyright by FPT 2018
18© Copyright by FPT 2018
A
B
Cable route from A to B drawn manually,
based on generic map
A
B
Efficient route from A to B generated by
machine learning, from collected data
Small lane from GPS data, not shown on generic map
Cross-road hanging from metadata