with Kubernetes and Tensorflow Daniel van der Ende & Tim van - - PowerPoint PPT Presentation

▶
with kubernetes and tensorflow
SMART_READER_LITE
LIVE PREVIEW

with Kubernetes and Tensorflow Daniel van der Ende & Tim van - - PowerPoint PPT Presentation

Aug 13, 2022 306 likes •886 views

Data Council NYC 2019 Reducing Flight Delays with Kubernetes and Tensorflow Daniel van der Ende & Tim van Cann o IT Consultancy o 40 Data Scientists, Machine Learning Engineers, and Data Engineers o Based in Amsterdam o We help

slide-1
SLIDE 1

Data Council NYC 2019

Reducing Flight Delays with Kubernetes and Tensorflow

Daniel van der Ende & Tim van Cann

slide-2
SLIDE 2
  • IT Consultancy
  • 40 Data Scientists, Machine Learning Engineers, and Data Engineers
  • Based in Amsterdam
  • We help organisations be successful with data and AI

Daniel van der Ende

Data Engineer danielvanderende@godatadriven.com

Tim van Cann

Data Engineer timvancann@godatadriven.com

slide-3
SLIDE 3
slide-4
SLIDE 4

499,444 499,444

Air Transport Movements at Schiphol

71.1 71.1

Million Passengers at Schiphol

79.2 79.2

Million Passengers

1.6 1.6

Billion Euro Real Estate

Mission Mission

Connecting the Netherlands

Ambition Ambition

Europe’s Preferred Airport

slide-5
SLIDE 5

What’s a turnaround?

slide-6
SLIDE 6

What’s a turnaround?

slide-7
SLIDE 7
slide-8
SLIDE 8

What’s a turnaround?

slide-9
SLIDE 9

More predictable Less predictable

slide-10
SLIDE 10
slide-11
SLIDE 11

Monitoring a turnaround

slide-12
SLIDE 12

High level Design

Magic Magic Eve Event nts

slide-13
SLIDE 13

High level Design

Magic Magic Eve Event nts

FuelingStart Fuel Fuelin ingEnd gEnd

slide-14
SLIDE 14

High level Design

Magic Magic Eve Event nts

slide-15
SLIDE 15

Streaming!

Unific Unification ation Object Object Det Detecti ection

  • n

Even Event t Gene Generation ration Even Events ts

slide-16
SLIDE 16

Streaming!

Unific Unification ation Object Object Det Detecti ection

  • n

Even Event t Gene Generation ration Even Events ts

slide-17
SLIDE 17

Unification of camera’s

slide-18
SLIDE 18

T0 T1

Unification of camera’s

T1 T1

slide-19
SLIDE 19

T0 T1

Unification of camera’s

T1 T1

slide-20
SLIDE 20

T0 T1

Unification of camera’s

T1 T1

slide-21
SLIDE 21

T1

Unification of camera’s

T0 T1 T1

slide-22
SLIDE 22

T1

Unification of camera’s

T1

slide-23
SLIDE 23

T1

Unification of camera’s

T3 T1

slide-24
SLIDE 24

T1

Unification of camera’s

T3 T6 T1

slide-25
SLIDE 25

T1

Unification of camera’s

T8 T3 T6 T1

slide-26
SLIDE 26

T1

Unification of camera’s

T8 T6 T3 T1

slide-27
SLIDE 27

T1

Unification of camera’s

T8 T6 T1

slide-28
SLIDE 28

T1

Unification of camera’s

T8 T6 T1

slide-29
SLIDE 29

T1

Unification of camera’s

T1

slide-30
SLIDE 30

Streaming!

Unific Unification ation Object Object Det Detecti ection

  • n

Even Event t Gene Generation ration Even Events ts

slide-31
SLIDE 31

Object Detection

slide-32
SLIDE 32

aircraft

  • pen_cargo_door

catering_truck conveyor_belt_loader fuel_tank_truck high_loader lavatory_truck pallet_transporter

  • pen_pax_door

fuel_pump_truck pushback_truck container

slide-33
SLIDE 33

"Batches are for Barbecues" Fokko Driesprong

slide-34
SLIDE 34

Object detection iterations

slide-35
SLIDE 35

Object detection iterations

1

slide-36
SLIDE 36

2

Object detection iterations

1

Serving Serving

slide-37
SLIDE 37

3 2

Object detection iterations

1

Serving Serving

slide-38
SLIDE 38
slide-39
SLIDE 39
slide-40
SLIDE 40

Iteration 1 Iteration 2 Iteration 3

CPU GPU

Object detection: GPUs or CPUs?

Approximate inference time per image(s)

slide-41
SLIDE 41

Streaming!

Unific Unification ation Object Object Det Detecti ection

  • n

Even Event t Gene Generation ration Even Events ts

slide-42
SLIDE 42

Why we need Event Generation

Object Object Detection Detection

{“objects”: [“aircraft”], “timestamp”: “2019-11-13 09:45:05”}

slide-43
SLIDE 43

Why we need Event Generation

Object Object Detection Detection

{“objects”: [“aircraft”], “timestamp”: “2019-11-13 09:45:05”} {“objects”: [“aircraft”], “timestamp”: “2019-11-13 09:45:10”} {“objects”: [“aircraft”], “timestamp”: “2019-11-13 09:45:15”} {“objects”: [“aircraft”], “timestamp”: “2019-11-13 09:45:20”} {“objects”: [“aircraft”], “timestamp”: “2019-11-13 09:45:25”} {“objects”: [“aircraft”], “timestamp”: “2019-11-13 09:45:30”} {“objects”: [“aircraft”], “timestamp”: “2019-11-13 09:45:35”} {“objects”: [“aircraft”], “timestamp”: “2019-11-13 09:45:40”}

slide-44
SLIDE 44

Event Generation

Aircraft present Aircraft Absent

slide-45
SLIDE 45

Event Generation

Aircraft present Aircraft Absent

slide-46
SLIDE 46

Event Generation

{ “event”: “AircraftArrives”, “timestamp”: “2019-11-13 09:43:25”, “ramp”: “X99”, “airport: “AMS” }

Aircraft present Aircraft Absent

slide-47
SLIDE 47

Event Generation

Aircraft present Aircraft Absent

slide-48
SLIDE 48

Event Generation

{ “event”: “AircraftDeparts”, “timestamp”: “2019-11-13 11:59:10”, “ramp”: “X99”, “airport: “AMS” }

Aircraft present Aircraft Absent

slide-49
SLIDE 49

if fuel_truck is present or aircraft is not present: return num_present = 0 for observation in window: if fuel_truck_detected: num_present += 1 num_present_ratio = num_present / window_size if num_present_ratio > presence_threshold_ratio: trigger_event(fuel_truck_arrives)

Event Generation “Business Rules”

slide-50
SLIDE 50

Other Dragons to Slay

Photo by Mateus Campos Felipe on Unsplash

slide-51
SLIDE 51

General Advice

slide-52
SLIDE 52

General Advice

slide-53
SLIDE 53

General Advice

Photo by Quino Al on Unsplash

slide-54
SLIDE 54

General Advice

Photo by Hunter Haley on Unsplash

slide-55
SLIDE 55

15:15 – 16:00 Room 568

(General) Advice is welcome

slide-56
SLIDE 56

Deep Turnaround Team

Aafke Jongsma

Product Owner aafke.jongsma@schiphol.nl

Contact

Daniel van der Ende

Data Engineer daniel.van.der.ende@schiphol.nl danielvanderende@godatadriven.com

Tim van Cann

Data Engineer tim.van.cann@schiphol.nl timvancann@godatadriven.com