Smart Maintenance Approach Using Data Analytics

Transportation is an essential service in the Indian economy and society. With an approximately 1.35 billion population, the Indian economy widely depends on the commute of its citizens. Indian railways act as a lifeline for the country, as it carries 13 million passengers each day. The passengers' safety is of supreme importance to the system. Over the years, besides the regular safety norms, the network has taken many steps through innovative technology use & stepping up its workforce training to improve safety standards. Some measures include replacing aging assets, advancing indigenous automatic train protection systems, and using the latest ICT (information & communication technologies) to make the Indian railway network smarter and enable more efficient maintenance.

 

Shifting Asset Management Trends

 

Indian Railways is encountering a shift in asset management structure, and efforts are in place to switch from periodic to condition-based and predictive maintenance. This will indeed help in lowering maintenance overheads on the present overloaded maintenance staff. Predictive maintenance techniques are created to help ascertain in-service equipment conditions to forecast when maintenance should be performed. This approach promises cost savings over routines or time-based preventive maintenance as the tasks are performed only when warranted.

 

The primary goal of predictive maintenance is to enable the convenient scheduling of corrective maintenance and reduce unexpected equipment failures. By knowing which tool needs maintenance, maintenance work can be better planned. Predictive maintenance has some benefits such as:

 

- It helps enhance the availability of signaling systems of lowering mean time to repair (MTTR) and boosting mean time between failure (MTBF).

- It reduces maintenance effort.

- It brings a collaborative work environment with OEMs (original equipment manufacturers).

- It helps make data-driven decisions.

 

The critical components of predictive maintenance include:

 

- Sensing the equipment's health parameters.

- Interacting the health data to a central location.

- Assessing this health data to render actionable decisions.

- Spreading the maintenance decisions to concerned investors.

 

Data analytics is the field where true value can be drawn. Various signaling assets are downloaded over the Indian railway network. Think of a track detection system. The systems for track detection range from traditional DC track circuits, created by W. Robinson in 1872, to modern axle counters, audio frequency track circuits, RFID tags, and balise. Likewise, the condition monitoring systems have become more complex and varied, and there is an entire sector working in this area.

 

Condition Monitoring For Assets

 

 

Condition monitoring systems for signals, point, integrated power supplies, axle, batteries, cables, interlocking systems, block instruments are needed to have successful condition monitoring for all signaling assets. Data loggers are already installed at many signaling installations. Condition monitoring systems can be synchronized with data loggers via RTUs (remote terminal units). These RTUs can be linked over a network to send all sensory data to a central location where the data warehouse is later created. All diagnostic and event data gathered from all the discrete monitored devices at a centralized location offers a chance to run multiple analytic app machine learning capabilities to render predictive maintenance schedules.

 

The data repository of diagnostic event associated information is the basis of predictive analytics. The more grainy data is made available, the more analytics algorithms can find trends, based on which the forecasts can be precise. The algorithms can crunch vast volumes of data to find trends and relations between various elements, like a relation between vibration & the mechanical wear & tear of a point machine's moving parts. Therefore, to have a brilliant and independent system, many sensors must be rendered to supervise the signaling asset's various parameters.

 

Since the data gathered from such sources will have variety, velocity, volume, and veracity, cloud platforms can significantly assess this data and get value out of it. Options like Amazon Web Services, Google Cloud Platform, MS Azure, and other products can be discovered to assess these datasets.

 

Data Analytics Before Condition Monitoring

 

 

Though these are projects in the pipeline to download a no. of sensors to supervise various parameters of signaling gears installed at the site, the data analytics power can be useful in finding asset failure trends based on historical failure as well. This way, the forecasts may not be 100% right, but absolutely render true info about the upcoming failures with a specific confidence interval.

 

For businesses that preserve huge asset numbers, it might not be possible to supervise every asset's condition in real-time. This needs ample funds, efforts, and time. Such investments can't be planned over a fortnight. There is always a likelihood of using past data failure to make estimates for the future. This can better plan maintenance activities.

 

For instance, the seasonal variation in failures. Assets like cables, track circuits, and some power tools are more prone to fail during spring when there is thunder, lightning, and rain, resulting in the waterlogging of tracks, etc., which ultimately leads to asset failures.

 

Past data failure can be examined to run regression models to find a relationship between the rainfall level at that time of year & the no. or sites of failures, etc. Likewise, other machine learning algorithms can help understand the potential of losing a specific asset in a given time frame if it has undergone a particular no. of cycles of operation.

 

These training models will require a substantial amount of data that has been organized precisely and adequately maintained overtime. ML models can help find a relationship between different variables & help us forecast the result with a certain level of exactness and probability.

 

An entire asset management system that uses sensory data & captures, supervises and assesses the asset condition data is undoubtedly the final goal to attain condition-based maintenance that lowers maintenance effort and lowers the entire lifecycle asset cost. However, for businesses in the previous stages of the transition process of captivating asset performance data on a real-time basis, past asset data can also be useful in lowering maintenance efforts by using correct data analytics tools. This is the big data power.