INDUSTRIAL asset management has traditionally viewed asset conditions in black and white, or rather in the red and green of a traffic light: either an asset is working, or it is not.

For example, a brake pad is either worn down and needs replacement, or it is fine for another maintenance cycle. In such a binary world, a seemingly sudden change from green to red can catch asset operators by surprise. To limit these nasty surprises, railway operators typically respond with significant - and very costly - preventive measures and excess capacity.

Hitachi Japan MaintenanceWith the advent of condition monitoring and diagnostic technologies - forming the field of predictive maintenance (PdM) - there are now fewer surprises. Different shades of yellow and orange emerge between red and green. For instance, with optical devices able to measure brake pad thickness during operation, it is possible to monitor and track the process of brake pad wear over time. Today, brake pad replacement has become less of a surprise for the operator.

Still, the PdM legacy technologies give insight and hindsight, but no foresight. PdM has a strong inspective and retrospective focus on post-event statistical evaluations and root cause analyses (RCA), with limited prospective capability. So far, very few operators have embraced forward-looking malfunction prognostics or asset availability forecasts.

Is PdM really a suitable basis for maintenance planning and life cycle management? Surprisingly, most operators are not asking themselves this question. Maintenance scheduling and scoping are still widely considered time-based and not condition-based.

Manufacturer (OEM) recommendations imply that schedules are set in stone, while established service contracts constrain operators. Other asset management decisions that go beyond maintenance follow the same rationale. Are operators going along with a generous maintenance calendar that serves OEM business interests, while their abundant - and highly useful - condition data is gathering dust?

Paradigm shift

Graduating from brake pad wear tracking to brake pad wear forecasting seems trivial. However, the move from "tracking" to "forecasting" is hardly that. New technology is bringing nothing less than a paradigm shift in the way operators think and talk about railway asset management, away from a focus on the past and present and towards the future.

Enhanced data collection, transfer, and analysis provide a wealth of intelligence that may soon become a game changer for railways. Big data, machine learning, the Internet of Things (IoT), and the Internet of Services (IoS) are not just technology buzzwords, but are instead finding their way into the industrial asset management arena.

For railways, in particular, these capabilities combine to enable Mobility 4.0, with its constant data generation and sharing between physical assets, humans, sophisticated analytical and simulation models, and "killer applications."

One of Mobility 4.0's killer applications is prognostics: the capability to take historical and current data and use it to create highly accurate long-term forecasts of future asset condition. In this way, prognostics answers the "when questions" of industrial asset management and has substantial implications for cost optimisation and downtime reduction.

Unlike legacy PdM technologies, prognostics goes far beyond insight and hindsight to provide foresight through novel forecasting technologies. This new breed of prognostic technologies is a translation of the stochastic process applications already found in finance, health care, and other industries. New prognostic applications let operators forecast asset condition and availability over time, test out various load and utilisation scenarios, and make informed decisions about optimal asset management, in particular on maintenance and replacement schedules.

Prognostics adds a new dimension to railway asset management: the ability to look into the future and forecast asset conditions, with implications for enhanced rolling stock availability. With this, operators are properly equipped to make decisions on optimal asset deployment, maintenance planning, and life cycle management.
The forecasting of asset conditions may address different levels of detail:

  • wear projection at the basic technical level
  • malfunction mitigation at an aggregate operational and commercial level, and
  • =xtension of remaining useful life (RUL) of assets at the strategic top level.

These forecasts can bring significant benefits for railways, given the high costs of preventive and reactive maintenance, and the stiff penalties and losses for delays and cancelled trains, mainly in Britain. Overall, prognostics minimises maintenance and downtime costs by offering:

  • suitable long-term scheduling and scoping of maintenance
  • maximised remaining useful life (RUL) through informed operations decisions, and
  • optimal deployment considering future asset risk profiles.

Although prognostics allows operators to answer previously unanswerable questions, it is a double-edged sword: increased transparency implies viewing maintenance needs as a function of risk. Prognoses are analyses of the probability of an asset being in a certain condition - an inefficiency, malfunction, or failure - at any point in the future. This is a paradigm shift from the deterministic view that railways have traditionally taken.

Within a railway company, it will take a significant shift of mindset – taking lessons from finance, health care, etc - to start viewing decisions as balancing known risk. Even with thorough precautions, risk is inevitable. Historically, companies have tried to minimise risk to acceptable levels without being able to actually consider the measure of that risk and how it develops over time. Prognostics allows this for the first time. Currently, money spent on redundancy and preemptive maintenance buys companies out of some risks, but nasty surprises like cancellations, delays, and unexpected maintenance remain common.


To take full advantage of prognostics, those within the organisation need to see risk as an opportunity rather than a necessary evil. With prognostics putting probabilities on risk, it is now possible to judge how acceptable various levels of risk are and take actions that - in the past - may have been deemed too risky. For example, redundancy can be reduced and maintenance made less costly if decision-makers judge that the risk of an individual wheelset reaching a parameter limit before the next scheduled maintenance is "acceptable," when in the past they would have brought it in for immediate maintenance.

However, balancing risk also requires balancing responsibility for risk-related decisions.

Traditionally, most asset malfunctions, defects, and breakdowns qualified as "bad luck." With a prognostic solution on critical assets, the risk of "bad luck" is a known basis for decision-making. A risk-literate decision-maker now has greater responsibilities, as fewer events are unforeseeable. Individuals will not be keen to own a decision-making process where the downside is greater without an associated upside for them. With that in mind, at what level in the organisation does risk literacy start? In addition, what is the upside for risk-literate railway decision-makers?

The risk-focused approach is burdened by a double paradigm shift: from a retrospective to a prospective paradigm, and within that from a deterministic to a risk-analytic paradigm. It is likely to receive some pushback from those on the frontline of maintenance and operation. Importantly, however, a change to a risk-analytic approach does not have to correspond to an increase in risk. Instead, risks that were previously unknown - creating a tendency towards both preemptive and emergency action - can now be carefully considered and reduced to a reasonable level, based on better operating and maintenance decisions.

In addition, prognostics does not remove experts and intuition from the equation. Operators' expert knowledge of correlation and causation is heavily involved in the forecasting process. Prognostics is, instead, an upgrade for railways in the era of Mobility 4.0. Vast quantities of data and sophisticated probability models supplement expert opinion to generate insights that could not come from either the data or expert opinion alone.


Predictive maintenance in railway operations

THE condition data that railways collect from their rolling stock includes thousands of data points, from laser measurements of single wheelset dimensions to vibration, temperature and lubricant parameter values on complex traction system. A variety of metrics are taken from critical components, and aggregated to specific vehicles, trains and entire fleets. This allows operators to spot condition changes and to assess their business impact rather quickly.

Monitoring is supported by diagnostics, which takes the raw condition data and allows operators to determine the exact locations in the complex system of specific problems such as inefficiencies, malfunctions and defects, and how they can be mitigated.

Both monitoring and diagnostics deal with the past and present of asset condition data. The latest trends in "predictive diagnostics" have further improved the detective power and sensitivity of diagnostic applications. Based on sophisticated pattern recognition and statistical inference, the latest diagnostic solutions can detect the earliest anomalies impacting industrial assets at any given time. "Predictive diagnostics" based on similarity-based models, in particular, compare empirical with simulated, ie "predicted" present conditions. This allows flagging of abnormal data trends as soon as they are statistically significant.

Operators are now exploring the prognostic capabilities of their data, beyond monitoring and diagnostics. Prognostic solutions enhance the long-term reliability and availability of rolling stock, and mitigate associated operational and commercial risk.