PREDICTIVE MAINTENANCE

We are excited to bring up the topic of Preditive Maintenance. Sometimes there are things in transportation that can be predicted before they occur, and analytics can help figure out what those specific things are. There are many benefits, approaches and real life examples of analytics at work in terms of Predictive Maintenance.

 

Benefits

Many transportation assets come with sophisticated and automated systems that can transmit information from
onboard computers and variety of other sensors. This has led to the new areas of research and analysis, referred to
as Asset Health Monitoring. There are locomotives and aircraft avionics capable of transmitting several thousand different sensory readings and diagnostics, almost on a continual basis. Many failures can now be prevented well
before they can cause any serious damage. Predictive maintenance is a highly desirable approach for reducing the
amount of unplanned maintenance and extending the life of components.

Many of the intelligent transportation assets are also equipped with the sensors to monitor the abuse by the
operators such as excessive speeding, acceleration, hard breaking, idling, etc. With appropriate analytics one can
improve the metrics such as asset life and fuel usage.

 

Analytical Approaches

The condition monitoring data is big in volume because of the sheer number of sensors present in an asset as well as because of frequency of measurements. Analyzing condition monitoring data from several assets to build a
predictive failure model is therefore a challenge. Condition monitoring readings are also prone to a high error rate due to the extreme operating environment. Hence, data validation and correction also plays an important role. Statistical methods of dimension reduction are used prior to building a predictive model. Advanced data mining techniques such as Association and Sequence analysis can be used to identify failures, which take place together as well as failure patterns where one failure is followed by other. The underlying theme is to look for “failure signatures” and to immediately take corrective action once they are detected. In terms of building predictive models, advanced data mining techniques such as decision trees and neural networks are employed.

 

Real Life Examples

One of the challenges in using the sensor information in trucking is the standardization. A large fleet owner may have assets sourced from multiple manufactures resulting in a need for standardizing the outputs form the sensors across the manufacturers prior to using for Analytics. Many trucks are now mounted with the telematics devices that transmit on-board diagnostics, location information and driver behavior information, including speed, duration at a stop, hard breaking, acceleration, fuel usage, etc. Analytics built using this information help improve the operational efficiency and reduce maintenance costs.
Some railcars have telematics devices mounted on them to determine the exact location. There is a recent
regulation that mandates these telematics devices on all tank cars carrying hazardous materials. Other than
telematics railcars generally do not have complex onboard sensors. Location of the railcars also can be identified
approximately using the Car Location Messages (CLMs), that are collected by the wayside CLM readers that identify the RFID tags on the railcars as they pass by.

Rail industry has an Equipment Health Management System (EHMS) that monitors the condition of the asset and
alert the owners when wheel repairs are needed. The sensors are not attached to the assets in this case; the system
takes readings of the railcars that go by using the wayside detectors. There are four types of detectors currently deployed in the North American rail infrastructure. They are Wheel Impact Load Detectors (WILD), Truck Hunting Detectors (THD), Acoustic Bearing Detectors (ABD), and Truck Performance (TP). Asset owners are slow to take advantage of this information during opportunistic level (giving them chance to schedule repairs before they start causing damage) to drive predictive maintenance. However, repair providers are actively using this information containing condemnable levels (telling shops that wheels need to be replaced) to actively attract business to their facilities.

Sensor data indicating aircraft operating conditions/health of critical components/systems is called AHM (Aircraft
Health Management). Usually AHM data is closely tied with unscheduled/preventive repair planning, where AHM data is monitored and any abnormalities lead to proactive scheduling of parts and repair on arrival. Various parameters are also monitored and analyzed to determine the health of engine. These include Engine Oil Analysis to determine the metal contents in the oil, Exit Gas Temperature analysis, where higher temperature of exiting gases indicates lower fuel efficiency of engines. This analysis is conducted to schedule engine washes that give a temporary improvement in performance, or fine-tune the repair work-scope when the engine is sent to the shop. The use of health/operating parameters data is more prevalent for engine maintenance as compared with the aircraft maintenance.

 

We hope to provide relevant insights and a fresh perspective on how analytics can make a profound impact on asset-intensive organizations. Stay tuned for our next post on Parts Inventory Management.