In the first blog in this series we explained Trenitalia’s challenges and motivations. This blog has more on current and future IIoT-enabled maintenance approaches the company uses, in transition from condition-based maintenance, via component-based maintenance to predictive maintenance.
IIoT-Enabled Maintenance Approaches
Trenitalia collects up to 10,000 parameters per locomotive each second, transmits these in real time via the Internet, and exploits them to better understand the health status of its fleet.
Current Real-time Condition-based Maintenance
Today, the company has close-to-real-time dashboards with equipment status information, events, and a sophisticated rules-based system incorporating expert knowledge to help decide between immediate or deferred maintenance. In some cases, a breakdown and reactive maintenance cannot be avoided. But in other cases, the company can react effectively to avoid breakdowns and the associated damage to its image.
If maintenance is required quickly, trains will most often complete their service. However, in rare cases, the company will send a new train to a station the damaged train can reach, put the new train alongside, and transfers the passengers. Losing five or ten minutes in the transfer is far better than getting stuck for hours in a tunnel!
Despite its benefits, the current system is still very close to reactive maintenance, and leads to inefficient interventions. According to Marco Caposciutti, CTO of Trenitalia, the remote diagnostic platform has limited scalability, anomaly detection is based on engineers’ experience only and limited by the number of installed sensors. Also, there is a lack of uniformity in how breakdowns are described.
Towards Predictive Maintenance
Since 2014, in close collaboration with SAP, the company brought those data into SAP HANA and started analyzing them. Francesco Mari, SAP Vice President IoT Business Innovation, noted that in this exploratory phase, all data are being gathered to determine which are useful. Mr. Mari explained that, for now, it is not possible to predict the breakdown of a component with high accuracy and precision.
Note: In predictive maintenance, “accuracy” expresses how often the model predicts a true positive compared to all occurrences. This number is often very high, for example over 90 percent. “Precision” in the same context is defined as true positives over predicted positives. This value can be much lower and is very hard to improve especially for complex assets as rolling stock.
However, the team identified more representative KPI’s than mileage. These include door opening/closing cycles. They distinguished groups of components with higher or lower risk. With this information, Trenitalia is transitioning to a dynamic, component-based maintenance strategy in which higher risk components and components reaching the limits of their KPIs are checked and maintained more frequently; while other components are checked and maintained less frequently. In some cases, diverging KPIs of components on the same train can be balanced by choosing specific destinations. For example, trips causing more left wheel rotations and accelerations can be balanced with destinations leading to more right wheel accelerations. Trenitalia had to make its integrated travel and maintenance schedules much more granular to achieve the desired massive increase in reliability and savings.
Danilo Gismondi, CIO of Trenitalia, mentioned that the investment in the new system is considerable: around €50 million, including tele-diagnostics and improvements in the maintenance plant. Mr. Gismondi said to be confident this investment will enable saving 8 to 10 percent of the current maintenance cost (€1.3 billion), increase availability by 5 to 8 percent, and reduce the cost of failures (€10 to €20 million) and associated impact on customers.
Technical Challenges Ahead
Although, Industrial IoT and analytics have simplified and speeded up the creation of value-adding applications, these haven’t made physics simpler. For example not all parameters can be measured on-line with sensors, amongst others at the outer surface of the train. Therefore Trenitalia has entered into a cooperation agreement with the University of Pisa to develop an automated tool to make inspections under the train when it stops in a station, and with the Politecnico University in Milan to develop soft sensors and algorithms for predictive analytics of fleet assets. We expect that all sensor, soft-sensor, and intermittent measures will be used as input for future predictive analytics. This is not to be taken lightly, since Trenitalia estimates that the data generated by train sensors will reach the petabyte range in 2018.
Ultimately, the system should provide predictions of breakdowns and maintenance needs, and send notifications to Trenitalia’s Dynamic Maintenance Management System, a solution based on the SAP Predictive Maintenance and Service solution. We expect this will complement, rather than replace, the dynamic component-based maintenance strategy. We will continue to follow this client case closely and report how the solution unfolds.
Culture Change For Company-wide Implementation
In a final blog in this series we will highlight the culture change required to make these new IIoT-enabled maintenance approaches a success.