tengizchevroil: ensuring safer motor vehicle operations
TRANSCRIPT
Tengizchevroil: Ensuring Safer Motor Vehicle Operations for a fleet of 5000
Daniel Barragan – TCO Business Intelligence Advisor
Bekmurat Spayev – TCO Data Scientist
2018 Tengizchevroil
Intelligence Motor Vehicle Safety Project
Objective
2
“World-class in process safety, personal safety & health, environment
impact, reliability and efficiency”
IT
Digitalization
Motor
Vehicle
Safety (MVS)
• Use technology to increase driver safety
• Encourage safe driver behavior
• Provide accurate and accessible information and data
Zero
Fatalities
TCO Key
Focus Area
Motor Vehicle SafetyTechnology Project
2018 Tengizchevroil
Intelligence Motor Vehicle Safety Project
Opportunities
3
To Improve Motor Vehicle Safety, Prevent Injuries, & Save Lives
Interactive Business Intelligence
• automated tools instead of manual reports
• geo-spatial visualizations
Timely Fleet Information and Reports
• shorten feedback loop to drivers
• uncover developing or underway issues
Real-Time Situational Awareness
• alerts & notifications
• real-time geo-spatial portal
Provide new insights
• violation detection
• traffic congestion
Predictive Capabilities & Analytics
• traffic congestion
• fatigue analytics
MVS Portal
• single view of most important data
• more immediate impact on safety
Road Conditions
• amber / red status
• quicker notification
2018 Tengizchevroil
Intelligence Motor Vehicle Safety Project
Roadmap
4
2017Phase 1 – 3Data Lake
Data Pump
June-July 2017Fatigue
ManagementProof of concept
2018 Ph4 W1BI Reports (5)
Events HeatmapsOvertaking
RCDB
2020 Ph4 W3Predictive Analytics
High Level Deliverables
2018 – 2019 Ph4 W2BI Reports (15)
Fleet ManagementNear RT VisualizationAccess Control and
Notifications
2018 Tengizchevroil
Intelligence Motor Vehicle Safety Project
Architecture
5
Query and update ETL Metadata (offsets, etc)
Stream data For NRT Analysis
...
Spark Direct Stream job per entity
...
SQL
Database
HDInsight
Spark
Key Vault
Query secrets
Event Hubs
Any data-science
toolkit
SQL Data
Warehouse (Data
Marts)
Microstrategy
Any geo-spatial
BI or mapsDetected overtaking events stream
HDInsight
Spark
Spark Direct Stream job forNRT events
Spark Direct Stream job per entity
...
HDInsight
Hive
...
Data is stored inORC/Tx Hive tables
...
Data Lake Store
HDInsight
Hive + LLAP
HDInsight
SparkEvent Hubs
event-overtaking
RedisStream to hot cache required for NRT Analysis data
Hot cache for NRT Analytics
HDInsight
Spark
HDInsight
Spark
Data
Sources
BI & Reporting
Ad-Hoc
Data
Lake
NRT Complex
Event Processing
NRT Machine Learning
Advanced Batch Analytics
Machine Learning
Batch Reports
SQL, JDBC, ODBC
Raw Area Prepared Area
Dedicated Hub for
ArcGIS GeoEvent Server
Legend:Telematics data flow
Internal infrastructure data flow
Detected violations data flow
External and ad-hoc data flow
Log Analytics
Log Analytics, Health Dashboards, Alerting
fleet
weather
road status
Notification
MVS Data Lake
External
Vendor
Data
2018 Tengizchevroil
Intelligence Motor Vehicle Safety Project
BI Architecture
6
AzureData Lake Store
ETLDevelopment(VIRTUAL MACHINE)
Data MartDevelopment
(AZURE SQL DATA WAREHOUSE)
BI AnalyticsDevelopment(VIRTUAL MACHINE)
PO
LYB
ASE
Data MartQA
(AZURE SQL DATA WAREHOUSE)
PO
LYB
ASE
BI AnalyticsUAT
(VIRTUAL MACHINE)
Data MartProduction
(AZURE SQL DATA WAREHOUSE)
PO
LYB
ASE
BI AnalyticsProduction
DATA MART
LANDING
DATA MART
LANDING
DATA MART
LANDING
ETLQA & Production(VIRTUAL MACHINE)
Component Description
Data Lake / Warehouse PaaS (Azure Data Lake Storage)
ETL IaaS (Azure Virtual Server, SSIS, Visual Studio in Dev only)
Data Mart PaaS (Azure SQL Data Warehouse)
BI Analytics IaaS (Azure Virtual Server, MicroStrategy)
ETL Packages Polybase (Warehouse to Landing), SSIS (Landing to Data Mart)
2018 Tengizchevroil
Intelligence Motor Vehicle Safety Project
Data Complexity
— Users
• 6,000 Vehicles (Will grow to 12k)
• 13,000 Active Drivers (Will grow to 20k)
• 140+ Contracting companies
• 500 Supervisors and Superintendents
— Data
• Constant ingestion of data from vendor
• > 115 M miles drives by vehicles in the last year
• 5.3M records added each day
• 3B records reflecting 3 years of historical data
• Data Mart is currently loaded 2x a day from lake
7
2018 Tengizchevroil
Intelligence Motor Vehicle Safety Project
Driver Fatigue, Distraction Detection and Intervention POC
— Guardian by Seeing Machines Ltd:
• Artificial Intelligence face tracking
system
— Proof of Concept Deployment Project
(June – July 2017):
• 60 days trial
• Representative sample of 50 vehicles
— Proof of Concept Results:
• Reduction in fatigue events – 96%
• Reduction in distraction events – 93%
— Next Steps:
• Scaled deployment to 450 vehicles in
2018, potentially further deployment to
1,050 vehicles in 2019
8
2018 Tengizchevroil
Intelligence Motor Vehicle Safety Project
BI Reports Wave 1: Event Heat Map Live Demo
14
2018 Tengizchevroil
Intelligence Motor Vehicle Safety Project
BI Reports Wave 2: Supervisor and Superintendent Dashboards
15
2018 Tengizchevroil
Intelligence Motor Vehicle Safety Project
BI Reports Wave 2: Supervisor and Superintendent Dashboards
16
2018 Tengizchevroil
Intelligence Motor Vehicle Safety Project
Overtaking Algorithm - Opportunity
17
Overtaking - act of one vehicle going past another slow moving vehicle, traveling in the same directions, on a road
- unsafe overtaking is dangerous- reported 5 to 10 unsafe overtakings per day in TCO
2018 Tengizchevroil
Intelligence Motor Vehicle Safety Project
Overtaking Algorithm - Solution
18
Layer Tools
Input - Vendor Telemetry Data
- GPS Point Positions Data
- Queue in the Azure Event Hub
Model - Spark Job
- Spark Java API
- Point of Interest and Vehicle Order Change
- 3rd Order Hermite Interpolation
Output - Output to the Azure Event Hub
- Stream consumed by 2 application
1. Overtaking Verification Application
2. Overtaking Persisted Jobs
2018 Tengizchevroil
Intelligence Motor Vehicle Safety Project
Overtaking Algorithm - Result
19
- 6000 suspected overtaking events per day
- 1500+ suspected unsafe events per day
- Validated – 70% positive violations
- First ever algorithm implemented in our fleet management system
- Next Steps:
- Discussion with business and stakeholders to figure out how we
will change drivers behaviors
- As accuracy and frequency of data improve the ways to increase
accuracy will be explored