ARTICLE CREDIT: ARC ADVISORY GROUP
ORIGINAL POST BY PETER REYNOLDS ON MARCH 31, 2016
Yousef-Abdul Moty, Director of Locomotive Engineering with CSX, a US-based rail transportation company, presented at the ARC Orlando conference, the company long term strategy and use of predictive maintenance, with Mtell as a solution provider.
Predictive Maintenance as a Business Strategy
In 2004, CSX changed business strategy to focus on profit drivers and the competitive advantage that could be achieved by enabling technology available in the industry. Corporate eyes quickly turned to asset utilization, since operating assets with the company exceeded $60 billion and mainly comprise of locomotives, cars, and track. The largest cost pool to address utilization were locomotives. Each locomotive was greater than $1 million in replacement cost and performance of the assets was showing 45 percent average availability. Like other companies or industries, infrastructure with CSX is aging. Some rail has the year 1902 stamped on the rails. Technology and innovation continue to expand, and the industry, regulatory and EPA continue to steer investment toward safety improvement.
Core CSX business is a transportation company, charged with safely, and reliably delivering customer goods on time, but plagued by operational and catastrophic failures that were undetected. CSX Reliability professionals knew the performance to failure curve all too well, and since 2003, CSX had experienced over $1 billion in total losses in 2500 train derailments due to human factors, track, locomotives and other causes.
CSX Turns to The Industrial Internet of Things
CSX started its own IIoT by deploying a series of interconnected sensors and wayside systems that continue study the car movements throughout the east coast of the US. CSX calls these “Super-sites” so as a locomotive and the entire train passes through these sites; the system reads the acoustic signatures on the bearings, peak impacts (pounding) on the rail (out-of-round wheels) and the geometry are looking at lateral forces such as wild or hunting trucks. IIoT, AEI sensors have the ability to count the axles to let the train crews know if there is an issue, how far back on the rail maintenance repair crews should begin.
IIoT delivered benefits by a drastic reduction in wheel derailments and bearing derailments due to the use of Predictive Analytics, IIoT and a shift to predictive maintenance. The most important part of the investigation for CSX was to focus on the identification of the point of degradation point “A” and preventing Operational Failure “B” and Catastrophic Failure “C”. Catch the failure before it happens and increasing the notice of the time to failure.
Core CSX business is a transportation company, safety, reliability and deliver customer goods on time but plagued by operational and catastrophic failures that were undetected. CSX Reliability professionals knew the Performance to failure curve all too well, and since 2003, CSX had experienced over $1 Billion in total losses in 2500 train derailments due to human factors, track, locomotives and other causes
In 2014, the company performed some analysis of more than a half-dozen solutions available in the industrial market and revealed a strong contender Mtell. Mtell was a leader in using machine learning technologies in conjunction with industrial assets. Mtell also invited CSX to visit TECK Mining in British Columbia. TECK mining was experiencing $1.5 million in avoided cost by using Mtell, which in turn helped CSX make the decision to pilot a solution with Mtell.
CSX Pilot Project to Predict Locomotive Engine Failures Oil Sample and Repair History
Two percent or 80 locomotives were failing each year and each asset costing over $100,000 each. CSX decided to embark on a three-phase pilot project, beginning with providing data to Mtell. CSX provided oil sample data to Mtell. Samples were random in nature, and each included a variety of wear metals, viscosity, and total base number. CSX had been using these samples to perform only simple threshold analysis, and alert mechanism and workflow. No multivariate analysis was done.
Pilot Phase: Mtell Find a Failure
The first portion of the pilot with Mtell involved allowing the company to look at existing data and find any possible failures or anomalies. Mtell was not made aware any relevant details of the samples of the CSX data and provided minimal explanation or backup material. CSX provided all of the most
frequent oil samples from 472 locomotives, of which Mtell chose nine samples to look closer. The nine samples contained a total of 22 data tags from oil and maintenance records together. This analysis and eight distinguishable patterns that were documented as failures. Shown below is the Mtell probability wave that provided a visualization of all 22 data elements shown together in one graph for easy interpretation. The red line showed an anomaly in the data. The event shown at the line created what CSX called a virtual work order to flag an event and look before and after the event.
The Mtell software revealed increases in iron, viscosity soot level and antiwear and the percentage each was impacting the overall signature. The Mtell software further identified 18 low water pressure problems on a particular locomotive, unaware that this locomotive was a bad actor. While CSX maintenance staff were fully aware of several cases of cooling water pressure failure causing crankcase bearing failure, they had not disclosed this to Mtell. This process was able to find and “train” the Mtell software agents against the failures and correlate this with maintenance history.
Pilot Phase 2: Predict Maintenance Failures
Once convinced by the initial data samples, CSX proceeded to look at a fleet of 235 locomotives, 8,635 samples, and 1,754 repair records. CSX then looked at 120 known failures (crank case overpressure, engine running hot, etc.) and built a model of the data with the known failures, marked the point and time and looked back in time, then provided to Mtell. Using this data, Mtell could have predicted the failures 60 days before the failure occurred. This model was then built on an active locomotive running fleet. On a specific locomotive, a failure was found during a transfer learning exercise. CSX found a pending failure by looking at the Mtell data. This flagged locomotive was routed to a maintenance facility for a pressure test on the cooling system. Normal maintenance pressure testing had missed the problem.
Pilot Phase 3: Prevent Failures
The CSX pilot extended to 600 locomotives. CSX has subsequently changed the way the operate, instead of running everything to failure, and performing preventative maintenance. CSX had not proactively changed out components. CSX created a new workflow called a Prescriptive Performance Improvement Request (PIR). Corporate Finance had previously run a financial analysis and decided proactive, predictive and prescriptive maintenance was not viable. OEM’s of equipment also dictated specific maintenance intervals to follow.
Several proactive maintenance failures have been identified and PIR’s created using the Mtell software including high crankcase wear, contamination (grit and sand) in the crankcase using oil samples and other data, making CSX believers in the pilot and taking the solution further. This original oil sample analysis phase brought all of the multi-variate analysis together and contextualized engine parameters like temperature, pressure, vibration.
The CSX Future for CSX Predictive Maintenance
In the near term, CSX is hosting Mtell internally, replicating agents and models to the entire fleet and integrating into the Maximo asset management system to automatically create PIRs within a Data Lake on SAP HANA and Hadoop. In the future, all of the communication systems from the locomotives will be integrated into Mtell running at corporate. Each locomotive has over 300 sensors or 1,200,000 data points for the enterprise.
Mechanical equipment specialists can perform system installation and configuration and manage all aspects of agent preparation and deployment including tasks to automate data auditing/cleansing, training agents with machine learning algorithms, and turning them into live monitoring implements that require little supervision.
ARC has been studying very closely the Emergence of Process Analytics Solutions that are capable of predicting maintenance, process anomalies and asset failures. Look for my report next month “The Curious Emergence of Process Analytics.”