MEASUREMENT TECHNOLOGIES AND MODELLING...
Transcript of MEASUREMENT TECHNOLOGIES AND MODELLING...
ALL TERRAIN SCANNING AND MODELLING
MEASUREMENT TECHNOLOGIES AND MODELLING METHODS
Gian Matteo Bianchi BSc MSc
Ph.D. student @ University of Rome - Tor Vergata
Principal Engineer Road Load Data @ Jaguar Land Rover
22 Nov 2018
Outline
The way to road scanning
2
• Pavement Surface Characteristics: PIARC categories and common road KPIs
• Measurement technologies
• Surface modelling: from point clouds to usable models, the CRG format
• A new approach to surface scanning: “an autonomous road scanning system”
The 4 Pillars of a CAE Simulation
Closing the loop between physical and virtual test
3
PAVEMENT SURFACE CHARACTERISTICS
INFLUENCE ON VEHICLE ATTRIBUTES
Texture Wavelength
An important concept
5
P. Pereira - Skid resistance and texture of compacted asphalt
mixes evaluated from the ifi in laboratory preparationISO13473-2:2002
λ
Pavement Surface Characteristics
Geometrical properties
6
1µm 10µm 100µm 1mm 10mm 100mm 1m 10m 100m
Microtexture Macrotexture Megatexture Roughness
Ride Quality
Wet Weather Friction
Dry Weather Friction
Vehicle Wear
In-Vehicle Noise
Texture
Wavelength
PIARC Category
Pavement
Surface
Characteristic
Influence
Tyre Wear
Rolling Resistance
Derived from XVIIIth World Road Congress in Brussel (Sept. 1987)
Pavement Surface Characteristics
Is it only civil engineering?
7
VTI - History of Road Profile Measurements
and Current Standards in Road Surface Characteristics
Beginning of XXth century
Highway Research Board bulletin -
Devices for recording and evaluating pavement roughness.
First mobile profilers – General Motors (1969)
TSL s.r.l. – High speed road profiler for macro-texture measurements (2013)
Pavement Surface Characteristics
A measure of comfort – International Roughness Index (IRI)
8
• The IRI is used worldwide by many
authorities
• An indicator of the pavement condition
• Measured using the road profile from 1.2 m
to 30 m (wavelength)
• Data is processed using a quarter-car “filter”
which represents the response of a standard
vehicle
• High IRI numbers mean that the road has
more imperfections, erosions,
depressions…worse ride quality!The Little book of Profiling – University of Michigan (1998)
Pavement Surface Characteristics
A measure of safety – International Friction Index (IFI)
9
• It’s a wet friction estimator
• The IFI quality is still debated
• Derived from internationally recognised
measurements: skid resistance and macro
texture (generally in the form of MPD)
𝐹 𝑆′ = 𝐹60𝑒(𝑆′−60)/𝑆𝑝
Function of
Macrotexture
Function of
both Macrotexture &
Skid Resistance
Pavement Surface Characteristics
A measure of safety – International Friction Index (IFI)
10
• The Mean Profile Depth: it’s an estimate of
the surface’s capability to drain water,
reducing risks of aquaplaning, spray and
splashes. ISO13473-1(2004)
• The Skid Resistance is directly measured
using two principles BS7941:
− Partially locked wheel, 15% slip ratio
− “Fix the vertical plan of the wheel at
20% to the line of chassis” (slip angle)
PK1PK2
𝑀𝑃𝐷 =𝑃𝐾1 + 𝑃𝐾2
2−𝑀𝑒𝑎𝑛
Mean
100 mm
Am
plit
ude [m
m]
Distance [mm]
Pavement Surface Characteristics
Skid resistance – measurement example
11
Std. Tarmac Filler Tarmac
Polished Tarmac
ParameterStd.
Tarmac
Filler
Tarmac
MPD [mm] 1.03 1.35
RMS [mm] 0.788 1.017
Standard Tarmac
Mean 0.75
Filler Tarmac
0.9
Polished Tarmac
0.8
𝑅𝑚𝑠 =1
𝑙 0
𝑙
𝑍2 𝑥 𝑑𝑥
Pavement Surface Characteristics
A relationship with efficiency – Rolling Resistance
12
Rolling resistance is mainly due to the (low frequency) hysteretic energy loss at the tyre-
road interface while the tyre deflects and flattens over the road surface
• Low rolling resistance tyres can reduce fuel consumptions by 1-2%
• It accounts for up to 30% of the usable energy* (based on speed)
• It is not only a property of the tyre but it’s due to the interaction between tyre and
surface
• The influence of this interaction is still debated, however independent studies have
proved that MPD and IRI can contribute to the RR up to 50%
*NHSTA Tire Fuel Efficiency Consumer Information Program Development Phase 2
MEASUREMENT TECHNOLOGIES FOR ROAD SCANNING
A BRIEF OVERVIEW
Light Detection and Ranging (LiDAR)
Back to basics
14
LASER emitter
Photodetector
𝜏 =2 ∗ 50 𝑚
3 ∗ 108 𝑚/𝑠= 300 𝑛𝑠
∆𝑑 = 10 𝑚𝑚 → ~0.07 𝑛𝑠
Time of Flight (ToF)
𝑑
Time
Counter
𝑑 =𝑐 ∆𝜏
2Resolution issue
∆𝜏
Am
plit
ud
e
Time
Light Detection and Ranging (LiDAR)
Back to basics
15
LASER emitter
Photodetector
Phase Shift
∆𝜑
𝑑𝑟𝑎𝑛𝑔𝑒1 =𝑐
2𝑓1=
𝑐
2 ∗ 2 𝑀𝐻𝑧= ~76 m
Ambiguity range based on modulation frequency
𝑑𝑟𝑎𝑛𝑔𝑒2 =𝑐
2𝑓2=
𝑐
2 ∗ 125 𝑀𝐻𝑧= ~1.2 m
𝑑 =∆𝜑 ∗ 𝑐
4 𝜋 𝑓
Phase detector
Am
plit
ud
e
Time
Light Detection and Ranging (LiDAR)
Market availability
16
What’s important?
− Angular Resolution --> Cartesian resolution
− RR and Accuracy --> quality
− Rotational Speed --> time!
− Degrees of Freedom --> 1 or 2, multibeam
Light Detection and Ranging (LiDAR)
Horizontal plan resolution
17
• Angular resolution is critical and it
follows trigonometry rules e.g.:
• 0.0072 deg equals ~25 mm point-to-
point resolution @ 20 m
• 0.0036 deg equals ~12 mm point-to-
point resolution @ 20 m
• Doubling the angular resolution
increases the acquisition time by a
factor of 4 (on 2 axis LiDAR)
Light Detection and Ranging (LiDAR)
Point density
18
• Angular resolution and sampling
distance give the point cloud density
• 2000 points / m2 is equivalent to a
point-to-point resolution of ~ 22 mm
• Reversing the math, a 5x5 mm grid
requires 40’000 points / m2
𝐷𝑒𝑛𝑠𝑖𝑡𝑦 =1
𝑅𝑒𝑠2
In the above example:Scanning area = 40 x 12 m2
Average density = 1460 points / m2
Average resolution = 26 x 26 mm
Total Number of points = 70 Million
Light Detection and Ranging (LiDAR)
Potential scenarios
19
Light Detection and Ranging (LiDAR)
Integration and application
20
Terrestrial
Laser Scanning (TLS)
Mobile
Laser Scanning (MLS)
Airborne
Laser Scanning (ALS)
Horizontal Resolution: 1 x 1 mm
Vertical Accuracy: 1 - 20 mm
Equipment Cost: Medium to High
Running Cost: High
Acquisition time: High
Horizontal Resolution: 5 x 5 mm
Vertical Accuracy: 10 - 50 mm
Equipment Cost: High
Running Cost: Medium
Acquisition time: Medium to Low
Horizontal Resolution: 30 x 30 mm
Vertical Accuracy: 20 - 60 mm
Equipment Cost: High
Running Cost: Low
Acquisition time: Medium to Low
Light Detection and Ranging (LiDAR)
Integration and application
21
Terrestrial
Laser Scanning (TLS)
Mobile
Laser Scanning (MLS)
Airborne
Laser Scanning (ALS)
Horizontal Resolution: 1 x 1 mm
Vertical Accuracy: 1 - 20 mm
Equipment Cost: Medium to High
Running Cost: High
Acquisition time: High
Horizontal Resolution: 5 x 5 mm
Vertical Accuracy: 10 - 50 mm
Equipment Cost: High
Running Cost: Medium
Acquisition time: Medium to Low
Horizontal Resolution: 30 x 30 mm
Vertical Accuracy: 20 - 60 mm
Equipment Cost: High
Running Cost: Low
Acquisition time: Medium to Low
Light Detection and Ranging (LiDAR)
Inertial reference – MLS and ALS
22
𝑥𝑝𝑦𝑝𝑧𝑝
= 𝑥 = 𝑥𝐼𝑀𝑈(𝒕) + 𝑀𝐼𝑀𝑈(𝒕)(𝑀𝐿𝑆𝑡𝑜𝐵 𝑟𝐿𝑆(𝒕) + 𝑟𝐿𝐴)
target Frame IMU position
in target frame
IMU orientation
in target frame
Sensor to body
frame orientation
Measured point
in sensor frame
Lever Arm
• GNSS + IMU accuracy can be as little
as 10 mm on x and y
• Z accuracy is limited by the angle at
which the receiver can see the
satellites, around 10-60 mm
• Body Roll, Pitch and Yaw are
important as well as position
• Boresight calibration is another source
of “static” errors
Light Detection and Ranging (LiDAR)
Can we see everything?
23
1µm 10µm 100µm 1mm 10mm 100mm 1m 10m 100m
Microtexture Macrotexture Megatexture Roughness
Ride Quality
Wet Weather Friction
Dry Weather Friction
Vehicle Wear
In-Vehicle Noise
Texture
Wavelength
PIARC Category
Pavement
Surface
Characteristic
Influence
Tyre Wear
Rolling Resistance
Airborne Laser Scanning
Mobile Laser Scanning
Terrestrial Laser Scanning
Texture Sensors
Back to basics
24
LASER emitter
Laser Triangulation
𝛼
baseline
Stereoscopy
baseline
Photogrammetry
𝑋𝐿 𝑋𝑅
Other technologies are available mainly for lab activities: induction, AFM, stylus, interferometer….
Texture Sensors
Laser triangulation
25
• Based on simple math and physics
• High speed > 60 KHz
• High RR and Accuracy but relative to the
measurement range
• Vertical resolution ~1 um
• Horizontal resolution limited by laser spot
size ~20-50 um
• It can be fitted on a motorised stage or on
a running vehicle….
• Target material optical properties may
create issues (specular or diffusive)!
• 2D variants can be adopted as well
ℎ =𝑏𝑎𝑠𝑒𝑙𝑖𝑛𝑒
tan 𝛼
𝛼
baseline
ℎ
Texture Sensors
Image correlation - principle
26
baseline
𝑋𝐿 𝑋𝑅
𝑧 =𝑓 ∗ 𝑏𝑎𝑠𝑒𝑙𝑖𝑛𝑒
𝑋𝐿 − 𝑋𝑅
𝑋𝐿 = ~1500 𝑋𝑅 = ~1750
Disparity
System setup
& calibration
𝐶1 𝐶2
Texture Sensors
Image correlation – disparity estimation
27
𝑋𝐿 = ~1500 𝑋𝑅 = ~1750
𝜌 = 0
“Block” Cross Correlation
𝜌 𝑑𝑖𝑠𝑝𝑎𝑟𝑖𝑡𝑦 =1
𝐾
𝑑𝑖𝑠𝑝𝑎𝑟𝑖𝑡𝑦=0
max _𝑑𝑖𝑠𝑝
𝐶1 𝑋𝐿 𝐶2 𝑋𝐿 + 𝑑𝑖𝑠𝑝𝑎𝑟𝑖𝑡𝑦
XR = XL − max( 𝜌 𝑑𝑖𝑠𝑝𝑎𝑟𝑖𝑡𝑦 )
𝜌 = 0.25
𝜌 = 1
𝜌 = 0.75
Very simplified approach, it’s a 2D problem!
Texture Sensors
Image correlation - limits
28
• Depth estimation from digital images
• Lots of math and estimations
• Quality (RR, resolution and accuracy)
depends on many factors: lens, camera
sensor, distance from object, setup…
• Horizontal resolution ~20-200 um
• Exposure is critical….
• Light source needs to be homogeneous
and controlled or results may be deeply
affected!
Balanced exposure level
Over-exposed
Texture Sensors
Image correlation
29
• Depth estimation from digital images
• Lots of math and estimations
• Quality (RR, resolution and accuracy)
depends on many factors: lens, camera
sensor, distance from object, setup…
• Horizontal resolution ~20-200 um
• Exposure is critical….
• Light source needs to be homogeneous
and controlled or results may be deeply
affected!
SURFACE MODELLING
FROM POINT CLOUDS TO USABLE MODELS, THE CRG FORMAT
Damper Response at Full Bump
As measured during a Road Load Data collection
31
Dam
per
Body F
orc
e [
kN
]
Vertical Suspension Travel [mm]
2.5 kN / mm
8.5 kN / mm
KDO 150 mm
Surface Modelling
Process workflow
32
Data Collection
(survey)
PCs
Registration
Complete PC
Cleaning
Meshing and
Processing
Usable Terrain
Model (CRG)
Geographical
Coordinates
and Metadata
CRG Library
Developed in-house
PC – Point Cloud
Surface Modelling
What is a point cloud (PC)?
33
It’s a collection of 3D scattered points
X = 1.251
Y = 0.215
Z = 0.535 m
Intensity = 100
Surface Modelling
Data collection (survey) – laser scanning
34
• Planning is crucial to get the right data
(resolution, missing info…)
• TLS requires a scan every ~5-20 m, time
consuming (based on resolution and surface
morphology)
• LS targets support processing and can be
used to import geo-graphical coordinates
• A LiDAR can see what you see,
line of sight is required
• Moreover, optical technologies don’t cope very
well with water/rain (scattering!)
Surface Modelling
Data collection (survey) – geographical information
35
• Geographical references can be imported in
the PC for post-processing
• Targets needs to be surveyed using
appropriate equipment called GNSS Rover
• Horizontal GNSS RTK accuracy is ~10mm
• Vertical GNSS RTK accuracy is ~20 mm
Lat. 66.0967221… deg
Long. 17.9830412… deg
Temp. -26 ºC
Surface Modelling
Point cloud registration
36
• It is necessary to build a unique PC starting
from different scan locations
• Each PC/scan location needs to be centred
and oriented to match one with the other
• 6 DoF (space and orientation)
• It’s power and time consuming
• This process can get to mean PC-to-PC
registration accuracies down to 0.8 mm
• The process is “similar” to the image cross-
correlation techniques and it can be facilitated
by physical targets
Surface Modelling
Point cloud registration
37
Surface Modelling
Point cloud cleaning
38
• It’s a manual process
• It takes time, based on the size and complexity
of the scenario
• It involves removing all undesired debris
• A leaf in a PC will be seen a “stone” in a CAE
simulation (surface deformation won’t be
modelled)
4 hours later….
Surface Modelling
Curved Regular Grid (CRG) format - OpenCRG
39
• It was created by Daimler AG (Germany)
• Highly efficient way to store the road geometry
• No need to store 3 coordinates for each point, like in
a Point Cloud (less data, faster processing)
• The surface trajectory is defined by the reference
line x(u), y(u), phi(u)
• The elevation z(u,v) is defined as a regular grid in
respect to the reference line position
• Data is stored in an un-curved regular grid
− u is the travelling direction
− v is transverse to the travelling direction
− z is queried by a unique u and v
− phi is the heading of the reference line
From OpenCRG User Manual
phi(u)
Surface Modelling
Curved Regular Grid (CRG) format - OpenCRG
40
From OpenCRG User Manual
• A constant spacing causes grid deformation on
corners, based on corner radius
• CRG can contain only static data in the Z matrix (e.g.
elevation, friction coefficient, temperature, grey
intensity….)
• The CRG format can’t be used to model any dynamic
behaviour in the tyre-surface interaction (e.g. speed
dependant friction levels)
• The CRG can store extra information \ metadata (e.g.
GPS coordinates, scaling factors, comments….)
• It’s open source!
Surface Modelling
The triangulation of a scattered point (cloud) set
41
• Many algorithms can be used to interpolate data over a
desired (regular) grid: triangulation, inverse distance,
radial, natural neighbour.....
• Delaunay triangulation is simple and effective
• It produces the largest minimal angle for each triangle
• The circle circumscribing any triangle contains no other
points of the original scattered PC
• It is commonly used in Finite Element Method (FEM)
• A generic point can be then extracted using interpolation
methods (linear, cubic….)
P1
P2
P3
P4
𝑧 = 𝑎𝑥 + 𝑏𝑦 + 𝑐
By 3 points there is one plane
Any point within the triangle
can be interpolated
Surface Modelling
Gridding\Meshing workflow
42
1. The triangulation is calculated
(heavy processing on a large dataset)
2. A regular grid is created, matching the required CRG
width and length
3. Each point (xn,yn) on the regular grid is interpolated on
the triangulation, estimating the zn
• Only points within the convex hull can be interpolated,
care needs to be provided while surveying\scanning
• Grid slicing and parallel processing can make this
process 10 times faster
• Time and computational complexity of meshing requires
𝑂(𝑟𝑒𝑠2) where 𝑟𝑒𝑠 is the regular grid resolution
Surface Modelling
From PC to CRG – visual example
43
Mesh size 10 mmOriginal PC
A NEW APPROACH TO SURFACE SCANNING
AN AUTONOMOUS ROAD SCANNING SYSTEM
An Autonomous Road Scanning System
Overview
45
• Measurements are taken in steady condition
(no inertial reference required)
• From 100 m down to 30 um
• Position and orientation is logged only to support post-
processing (geographical information)
• Path follower, autonomous driving based on a known
plan
• Max speed 1 m/s (currently)
• Same acquisition time as TLS but extra information is
acquired (micro-texture and track temperature)
• Patent Issued on Feb 2018
An Autonomous Road Scanning System
System integration
46
• Developed in Robotic Operating System (ROS)
• Wireless capabilities up to 1 km (LoS)
• Safety features include: watchdogs, emergency braking
and remote kill switch
• Comms: Wi-Fi, eth, CAN, serial
• Ground station developed in-hose
An Autonomous Road Scanning System
Results
47
• Given a plan (track or grid), the robotic platform follows
the path, stopping at each measurement point while
driving along waypoints
• User sets spacing between TLS, texture and
temperature measurements
• The ground station provides a feedback on system
status and control parameters
• Suitable for proving grounds and confined test areas
Track Temperature = 18 °C
An Autonomous Road Scanning System
Toward friction estimation
48
1µm 10µm 100µm 1mm 10mm 100mm 1m 10m 100m
Microtexture Macrotexture Megatexture Roughness
Ride Quality
Wet Weather Friction
Dry Weather Friction
Vehicle Wear
In-Vehicle Noise
Texture
Wavelength
PIARC Category
Pavement
Surface
Characteristic
Influence
Tyre Wear
Rolling Resistance
Airborne Laser Scanning
Mobile Laser Scanning
Terrestrial Laser Scanning
JLR – Autonomous Road Scanning System
Almost 2 orders of magnitude
Jaguar Land RoverW/1/26 Abbey Road, WhitleyCoventry CV3 4LF, UK
jaguarlandrover.com
THANK YOUGrazie!
Gian Matteo Bianchi BSc MSc PhD StudentPrincipal Engineer Road Load Data
M +44(0)7469 416805
49
University of Rome – Tor VergataVia del Politecnico 100133 Rome, Italy
ing.uniroma2.it