Railroads
TNP’s AI-MSET Prognostic Innovations for Railroad Predictive and Prescriptive Maintenance, and for Avoidance of Catastrophic Derailments; Costly Downtime; Service Interruptions
Based on the world's most effective and proven AI Anomaly Detection algorithm, True North Prognostics (TNP) has developed a predictive maintenance framework for Railroad locomotives, rail cars (freight and passenger), and Switch-and-Crossing (S&C) systems and their embedded safety-critical instrumentation.
TNP’s real-time surveillance of on-board sensor telemetry (temperatures, voltages, RPMs, currents, tri-axial vibration sensors, diesel engine telemetry) and wayside sensors that can include vibration, acoustics, and IR-thermal imagery, provides early-warning of incipient anomalies in wheels, roller-bearings, axle journal integrity, as well as diesel engine mechanical and thermal-hydraulic integrity. Early warning prognostics from TNP’s AI framework, called AI-MSET, enables tiny developing faults to be proactively remediated before the defects grow and lead to derailments or costly downtime and service interruptions.
AI-MSET is a 3rd generation of a mature and well proven AI Machine Learning (AI/ML) technique called the Multivariate State Estimation Technique (MSET), which was originally developed by the USDOE’s Argonne National Laboratory in the 1990’s for monitoring instrumentation in commercial nuclear reactors. MSET1 was spun off for prognostic applications aviation, aerospace, precision manufacturing, US Navy propulsion and weaponry prognostics, data center business-critical IT assets, Disney theme parks worldwide for all their park rides structural and mechanical rail-cart sophisticated real-time safety-assurance systems, and for both EMD and GE locomotive early-warning anomaly detection.
For Railroad predictive and prescriptive maintenance use cases, AI-MSET’s advantages over competitive forms of AI/ML include:
1. Early warnings of incipient anomalies, days and often weeks in advance of discovery by competitive monitoring approaches
2. Lowest possible False-Alarm and Missed-Alarm Probabilities (FAPs/MAPs)
- FAPs are extremely costly for Railroads … taking revenue-generating assets out of service unnecessarily
- MAPs lead to costly downtime and catastrophic derailments
3. Highest computational scalability, which yields orders-of-magnitude lower compute costs versus competitive Neural Network anomaly detection
- AI-MSET scales to PetaByte-per-day sensor volumes generated by large railroad companies
- Sophisticated real-time AI prognostics for sensors with high frequency waveforms (vibrations, acoustics, IR-Thermal-Imagery)
4. Fast, accurate Traceback Root Cause Analyses
- Competitive neural networks cannot do root cause analyses (RCAs)
- In the US, train derailments can take over a year of NTSB investigations to establish RCA
- Because TNP’s AI-MSET is deterministic math, Railroad Services engineers can immediately (within 48 hours) perform fast, accurate, auditable, and transparent RCA
- Important for Insurance Agencies and Govt Regulator Agencies – establishing whether events were due to: Manufacturing defects, maintenance negligence, “force de majeure” acts of nature (e.g. lightning strikes, tornadoes, etc.), human operator error, malicious sabotage, etc.
TNP’s Advanced Pattern Recognition software (called AI-MSET) will provide enhanced rail safety, early detection of incipient degradation in rail bearings, wheels, axles, diesel systems, and also early detection of developing problems in the tracks (when 2 or more trains using the same tracks start to show temporary incipient anomalous patterns while passing the same GPS location, a GPS-tagged Service Request is generated for crews to examine the track integrity at that exact latitude/longitude location).
TNP’s AI-MSET algorithmics provide enhanced predictive and prescriptive maintenance that will address (and mitigate/avoid) the predominant sources of derailment incidents and catastrophic faults in locomotives and rolling stock (both freight and passenger cars), and in critically important Switch-and-Crossing systems and associated instrumentation.
Advantage for countries that may have aging legacy railroad systems: [both trains and tracks]
TNP AI Algorithmics works with any legacy railway systems, and any vintage of sensors [on-board and wayside]. This means that the Government Minister of Transportation (for nationalized train systems) and/or the leaders of Private Railroad Systems can announce that there will be very quick short-term benefits to passengers and freight shippers. All TNP AI algorithms are implemented in Software and require no additional hardware or modifications to quickly see a quantifiable reduction in failure probabilities and scheduling delays from unplanned downtime. Thus, even for commitments to upgrade locomotive and rolling stock investments [which might be spread over months and coming years], there will immediately be higher safety margins, much lower Operations and Maintenance (O&M) costs, and much fewer scheduling delays for passenger and freight customers when trains have unanticipated breakdowns in service.