Development and Demonstration of the Truck Activity Monitoring System (TAMS)

ARB Research Seminar July 13, 2016, 1:30 PM PDT

Presented by:

Stephen G. Ritchie, Ph.D. Andre Y.C. Tok, Ph.D. Director Asst. Project Scientist

Institute of Transportation Studies University of California, Irvine

Data Groundtruth System

2

3

Model Design Two Systems of Body Classification Models Developed:

Inductive Signature only Model

(for existing ILD sites)

Integrated WIM and Inductive Signature Model

(for existing WIM sites)

1st to 2nd 3rd to 4th 4th to 5th2nd to 3rdAxle Spacing

WIM Site Data

Inductive Signature Dat a

Axle Weights + Axle Count, Gross Vehicle Weight, Vehicle Length

Inductive Signature Dat a

Duration

Body Classification Architecture: Two Systems of Models Signature Only Model System WIM and Signature Model System

4

Tier

1Ti

er 2

Tier

3

Single-Units Multi-Units

Signature Data

13 Classes

8 Classes

19 Classes

7 Classes

Single Unit Truck

Single Unit w/ Trailer

Tractor w/ Semi Trailer

Multiple Trailers

Passenger Vehicle

Paired WIM and Signature Data

FHWA Class 4

FHWAClass 5 - 13

FHWA Class 14

4 classes

Trailer Detection*

10 to 16 classes 5 classes

FHWAClass 5 - 13 FHWA

Class 5 - 13 FHWA

Class 5 - 13

Signature Only Model Results

• Body class model results summary – 4 categories incorporating 47 truck body classes – 34 classes with classification accuracy > 70% – 27 classes with volume error < 10%

Sub-Model Classes Accuracy (%) Volume Error (%)

Passenger Vehicles 1

Single Unit Trucks 13 72.3 15.4

Single Unit Trucks w/ Trailer 8 94.2 8.2

Tractors with Semi-Trailer 19 74.3 11.3

Tractors with Multiple Trailers 7 90.4 7.0

Signature Only Model Results (for Semi-Tractor Trailers)

Vans with 5 or more axles Reefer Vans with 5 or more axles

Vans with less than 5 axles Reefer Vans with less than 5 axles

53ft Container 40ft Container

40ft Container Reefer 20ft Container

Platform Tank

Open Top Van Auto

Low Boy Platform Drop Frame Van

Dump Logging

Livestock Agriculture

Beverage

FHWA 8

FHWA 9

FHWA 10

Overall 74.3% Correct Classification Rate across

19 body classes

6

75% 74%

83% 46%

57% 75%

94% 86%

78% 71%

61% 78%

82% 61%

70% 80% 80%

59% 93%

40% 50% 60% 70% 80% 90% 100% Correct Classification Rate

Integrated (WIM + Signature) Model Results

- System of 9 sub-models with 63 body classes - 52 classes with classification accuracy > 70% - 37 classes with volume error < 10%

Accuracy Volume Error Model Classes (%) (%)

FHWA 4 4 95.2 9.7 FHWA 5 10 75.3 6.8 FHWA 6 8 80.5 9.2 FHWA 7 4 100.0 0.0 FHWA 8 5 90.9 4.2

FHWA 9 Semi Tractors 16 75.4 12.2 FHWA 9 Single Trailers 5 96.7 1.7

FHWA 10 4 92.3 7.7 FHWA 11 and 12 7 92.7 8.0

7

Integrated (WIM + Signature) Model Results

FHWA Class 9

Overall 76% Correct Classification Rate

16 body classes

Enclosed Van

Reefer Vans

53ft Container

40ft Container

40ft Reefer Container

20ft Container

Platforms

Tank

Open Top Van

Dump

Auto

Low Boy Platform

Drop Frame Van

Logging

Livestock

Agricultural Van

72% 75%

55% 82%

94% 100%

81% 78%

80% 67%

94% 89%

74% 86%

96% 64%

50% 60% 70% 80% 90% 100% Correct Classification Rate (CCR)

8

Hardware Components

Advanced Detector Cards

(Acquire Inductive Signature Data)

Fanless Field Computer

(On Site Data Processing)

Wireless Modem (Communications

to Server)

9

10

Types of Site Deployments

Inductive WIM-Signature Signature Only Integration

Traffic Monitoring Site on CA-4 near Stockton

Ramp Metering ILD Site

on SR-91 in LA

WIM Site on SR-99 between

Stockton and Sacramento

Census AVC Site on I-15 in Escondido

Data Flow Architecture Overview

Weigh-In-Motion Site UCI ITS

WIM Processing

Signature Processing

Signature Detector

Cards

SSH Tunnel

WIM Controller

Channel 1

Channel 2

Channel 3

Channel n

Field Unit

REDIS Middleware

Signature Processing

Signature Detector

Cards

SSH Tunnel Field Unit

Database Bridge

PostgreSQL Database

Inductive Loop Detector Site

11

LIVE DEMO

1. Live Signature Transmission 2. Walk Through TAMS Web Interface

12

Some Collateral Benefits

• i.e. PierPass: Monitoring truck port activities Policy Evaluation

• Monitor truck lane violations • Monitor unauthorized travel along restricted routes Enforcement

• Determine the impacts of port strikes, freeway closures, etc.

Impact Assessment of Non-recurrent Events

• Ability to analyze temporal and seasonal variations of truck impacts by industry

Understand Industry Impacts on Traffic, Infrastructure and

Emissions

• Through integration with anonymous signature-based truck tracking and/or survey studies

Improved Truck VMT Estimates

13

Just Getting Started: A platform for future research and applications

• Raw signature data is being continuously archived

• Case in point: Activity inventory of cleaner trucks – Recognize trucks by model

year to obtain activity trends of clean vs older gross-polluting trucks?

14

A Work in Progress: Improvements are ongoing…

Increase robustness of field deployments

Improve classification accuracy of trucks

Implement classification models for state and

federal agency reporting requirements

15

QUESTIONS

Stephen G. Ritchie sritchie@uci.edu

Andre Y.C. Tok ytok@uci.edu

Truck Activity Monitoring System, http://freight.its.uci.edu/tams

16

Model Development

17

Redding (SB I-5)

Willows (NB I-5)

Fresno (SB SR-99)

Irvine (SB I-5)

Data Collection Sites • Land use variation: four sites with

differing land use characteristics • Comprehensive data: still image photos,

WIM data, and inductive signatures • Temporal variation: multiple times of

day, days of week, and seasons included

• 97 hours of data collected, with 35,000 vehicle records (mostly trucks) processed for model development and testing

18

Presence actuated Digital still image camera

Inductive Signature Detector Cards

WIM Controller

Data collection setup at Fresno WIM site

19

Where it all began… Investigation of Blade Inductive Sensors

Installation of Blade Sensors

• 1st vehicle classification study tofocus primarily on distinguishingdetailed truck configurations

• Utilized a temporary surface mount sensor

• Model capable of identifying over100 truck configurations

Truck crossing inductive sensors

Sensor layout at freeway exit ramp to San Onofre Weigh and Inspection Facility

Double Blade Sensors

Preformed 6’ Round Loop Sensor

20

Background

• Improved Validation and Calibration of the California Statewide Freight Forecasting Model (CSFFM)

• Enhancement of truck classification models • Expand deployment to over 90 locations along major

truck corridors across California, encompassing – state borders, – regional cordons, and – metropolitan areas

Pilot study funded by CARB in 2012

Current study funded by CALTRANS in 2015

• Initial development of inductive signature-based truck body classification models

• Deployed at 16 weigh-in-motion (WIM) and inductive loop detector (ILD) sites in the California San Joaquin Valley

What is the Truck Activity Monitoring System (TAMS)? A truck counting system that is…

21

Spatially Representative –Will be deployed at over 90

major truck corridors across the State of California

Temporally Continuous –Data collected and

transmitted real-time 24/7

Sustainable –Leverages existing Inductive

Loop and Weigh-In-Motion Detector infrastructure

High Fidelity –Identifies 40 to 60 truck / trailer

body configurations

Advanced –Adopts Inductive Loop

Signature technology (combined with Weigh-In-Motion technology where available)

Accessible and Automated –Hosted on an interactive GIS-

enabled web-based user interface

Potential Applications

22

Estimate proportions of freight and non-freight truck movements

Statistics relating to empty movements in freight trucks

Temporal and spatial travel behavior of trucks by industry

Estimate proportions of long and short haul trips along major and

restricted truck corridors

Better understanding of truck travel patterns and

behavior

Detector Technologies Behind TAMS

Two Types of Detector Solutions:

Combined Weigh-In-Motion (WIM) and Inductive Signature

Technology at existing WIM sites

Standalone Inductive Signature

Technology at existing Inductive Loop

Detector sites

23

Weigh-In-Motion Technology Components

• Bending Plates – Measure Wheel/Axle

Weights

Bending Plates

Provides 13 axle-based FHWA classifications (14 in California)

Inductive Loop Sensors Traveled lane on freeway

• Inductive Loop Sensors – Presence detection

Over 100 Data WIM sites in California located along Provide speed, vehicle Major Truck Corridors

length and axle spacing measurements Weigh-In-Motion sensors located

along a freeway

Inductive Signature Technology

Time In

duct

ance

Cha

nge

Time

Indu

ctan

ce C

hang

e

Conventional Measurement [0,1] Binary output typically sampled at 30 samples/sec

Inductive Signature High resolution inductive magnitude changes at up to 1000 samples/sec

• Conventional Inductive Loop Detectors (ILDs) produce bivalent outputs – Generate traffic counts, not truck counts

• Advanced ILDs measure inductance changes ‘Inductive Signature’ – Inductive signatures are indicative of body configuration

25

Sample FHWA Class 9 (5- Axle Semi-Trailer) signatures by trailer configuration

Enclosed Van Livestock Low Boy Platform

Tanks Drop Frame Van Basic Platform

26

Inductive Vehicle Signature Applications

27

Real-time Section Travel Time and Speeds

Section-level Density Section-level Emissions Estimation

Comparison of "True" and Estimated 30-sec Aggregated Average Speeds during PM Peak of 06/08/2009

on NBI-405 @ Sand Canyon (Model Trained from SB I-5 @ Crescent) 80

70

60

50

40

30

20

10

0 15:00 16:00 17:00 18:00 19:00 20:00

Adjusted True Speed Estimated Speed Residual

Single Loop Point Speed Estimation

Truck Configurations

,...... ---

Bulk waste

Trailer Configurations

30

Existing Truck Activity Data Sources

• GPS • Cellphone

• Weigh-In-Motion • Automatic Vehicle

Classifier (AVC) System

• 2002 National Vehicle Inventory and Use Survey (VIUS)

• 2016 California VIUS • Regional Intercept

Surveys

Mobile Static / Count Data Surveys

31

Research Objective

Develop and deploy cutting-edge classification models that

– leverage existing detector infrastructure – to provide truck count data by detailed

configuration – using advanced inductive signature technology