LHP Data Analytics
Engine Diagnostics: Emissions Fault Analysis
Executive Summary
Michael King
President, Data Analytics & IoT
LHP Engineering Solutions
http://LHPES.com
Engine Diagnostics Analysis: Background
• Cummins Emission Solutions (CES) focuses on the design and development of aftertreatment systems designed to meet stringent emissions requirements as well as customers expectations for high quality.
• In preparation for the 2017 product launch, CES requires reactive and proactive monitoring of system performance to identify trends or characteristics of performance so true failures can be addressed before they occur.
• One such data source that may be underutilized to identify trends in performance is engine ECM images captured by the Cummins INSITE service tool during each service event. These images are then stored in a Cummins database.
• CES wishes to explore the ‘art of the possible’ regarding what trends can be identified by interrogating this database on a pilot basis through the services of experts in LHP Data Analytics.
Engine Diagnostics Analysis:
Problem Statement
1. Tell us [Cummins] what we already know
2. Provide Fault Code based analysis of historical INSITE images
3. Provide expertise around opportunity and how to use Machine Learning with Cummins’ data
4. Provide future opportunities, analytical techniques, proactive analysis, and support for next steps
Phase I: Develop Baseline Data Analytics
• 2016 ISX Engines– INSITE Images
• Faults & Parameters– 1682/5655 and 2771 Faults and Fault Parameters
– 1682/5655 and 2771 Aftertreatment History Parameters
Phase II: Establish Reactive, Proactive, and Predictive Analytics
• Expand engine population to include 2017 test
• Incorporate Data Loggers, manufacturing data, and additional data sources
Phase III: Deploy “Gold Standard” Analytics for 2017 Products
• Global deployment of common analytics
• Establish training and support plan
Engine Diagnostics Analysis:
Project Scope
• Leveraged existing Cummins ECM INSITE image database
• Developed Fault-to-Analyze cross reference
• Established Parameter cross references
• Developed Engine Hours logic across the data set
• Calculated First Time to Failure
• Performed Lead/Lag analysis
• Deployed R graphical capabilities for advanced analysis
• Future logic enhancements– Nature of generating an INSITE image
– Correlate odometer reading to service bay mileage reading to see how long an operator drove with the fault indicator
– Additional value added data sets and their measures
LHP Data Analytics Methodology
• Mean time to First Fault and All Faults are increasing
– Average of 2.98 days to first Fault
– Average of 6.50 days to first SCR Fault
• General trend is improving for % Engines with Faults, Days to First Failure
• Potential issues related to:
• Kenworth Mexico
• MDC Shipments
• ISX2 2013
• Software Calibration Phase
Engine Diagnostics Analysis:
Initial Results
• Incorporate Data Loggers to provide real-time data
– Expand Machine Learning and Predictive Analytics
– Provide geographic and climate location
• Expand engine data population, incorporate manufactured data
– Epidemiology: Tie to part failures / manufacturing failures / lot numbers
– Expanded pre-Fault Lead/Lag analysis
• Integrated Marketing and Warranty programs
– ESN to VIN to customer for advanced notifications for engines at risk
– Validate INSITE images against claims (three level match)
• Incorporate ECM hardware and software versioning
– When the ECM was flashed, which version
– Tie to any software issues
• Engineering thresholds and real time data
– Abort Fault conditions, operating ranges
Engine Diagnostics Analysis:
Next Steps
Highlighted Engine Diagnostics Analysis
Aftermarket Faults
• Mean time to initial Faults is high after initial launch– Average of 2.98 days to first fault
– Average of 6.50 days to first SCR fault
– General trend improves after 6 months:• % Engines with Faults
• Days to First Failure
Aftermarket Faults
• Overall trends are improving for 2016 ISX
• Analysis identified Potential issues related to• Kenworth Mexico
• ISX2 2013
• Deeper dive into Kenworth Mexico to see their issues and why they were an outlier:• Issues early in engine
build/release
• Issues starting to reappear in October
Kenworth Mexico
Aftermarket Faults Selected
Fault To Analyze = 2771
• Two undesirable insights
• Lots of Faults for engines shipped to MDC
• MDC engines also have high severity Faults
• Identification of lead/lag analysis for preceding Faults
Fault To Analyze = 2771
Parameter Analysis
• Several parameters indicate failures occur:
• Early in lifecycle
• At low temperature thresholds
Calibration Software Deep Dive
• Isolating Calibration Software root cause to failure:• Specific version of
Calibration Software has significantly more Faults as a % of their installed population than others
Engine Diagnostics: Executive Dashboard
Engine Diagnostics: Fault Snapshot
Engine Diagnostics: Lead-Lag Analysis
Engine Diagnostics: Fault Code Deep Dive
Machine Learning Capability
• Identifies key parameters that are leading to specific Fault Codes
• Mutual Information identified unexpected potential variables to analyze
• Provides rapid Feature selection and down selecting or reducing data
• Final parameter list to investigate in more detail
LHP Data Analytics Solutions
• Technical and Analytics
• James Roberts
• Vice President, Data Analytics
Solutions
• 812.314.7921
• Michael King
• President, Data Analytics
Solutions
• 812.341.8460
• Account Management
• Paul Wright
• Director, Business Development
• 812.314.7920