Asset Management & Smart Maintenance

Prognostics: Improving decisions in asset management through data-driven strategies – by Moritz von Plate

Condition Monitoring is a good basis for asset management – but it’s only the beginning

Every day people make decisions: We choose to make an appointment at the dentist on Monday or to have our car inspected on Tuesday. At work, however, people are frequently confronted with more complex issues. Especially in the industrial sector, a single mistake can cause enormous costs. Among professionals confronted with decisions that allow no mistake on a daily basis are asset managers generally and maintenance managers specifically. The good news is that there are tools that can help make better decisions based on smart data analytics instead of only experience and intuition. While Condition Monitoring is now established as a decision support tool, there are new technologies emerging that can predict explicit future time horizons for interventions and thus allowing decisions that incorporate explicit information about the future.

Technology will improve decision making

During past decades various technologies for condition monitoring have been established as best practice. Application and effective use of the individual technologies depend, among other things, a great deal on the type of equipment at hand. Among the popular ones used for equipment with fast rotating elements, such as turbines, pumps and compressors, are vibration analysis, temperature monitoring and lubrication analysis (tribology). Machine operators and their technical service providers recognize critical operational conditions in advance as a warning is triggered. A diagnosis, usually conducted by human experts, sometimes complemented by algorithms, determines imminent technical malfunctions and the reasons behind them. This normally results in specific measures to fix the malfunction. A suitable time-frame for such measures is determined through planning tools.

When it comes to condition monitoring, a warning or alarm triggered by condition data or a detailed condition diagnosis is generally more valuable the earlier it indicates a critical operational state. Even more valuable is the prognosis about the explicit future time horizon left until the malfunction is likely to occur. Without such an explicit forecast horizon condition-based maintenance is not really practical, even possible. Before the introduction of Prognostics most equipment operators had to work through preventive or even reactive maintenance, complemented by alarm, diagnosis and intervention functions with a relatively short advance warning time. Still today it is often mistakenly believed that data collected through condition monitoring is not sufficient to deliver a useful basis for prognosis. But in fact it is and in light of the relatively large technical effort already carried out to collect, process and archive data as well as the availability of prognostic data analysis solutions, it is surprising that the opportunity of creating an explicit prognostic horizon is not yet used by all operators.

Application for better decision making

By using a computational stochastic process model and the empirical data base compiled by condition monitoring tools, prognostic reports can be generated that form the basis for optimized maintenance decisions. For example at Cassantec we have developed a tool that provides: consolidation of condition data, condition forecasts and related malfunctions prognoses, including a standardized prognostic horizon. Additionally, our solution offers a comprehensive report of the Remaining Useful Life (RUL) distribution and an automated periodical update of the results. The condition monitoring functions such as alarming and diagnosis are complemented by an explicit prognostic dimension. The resulting foresight can be used for both mid- and long-term maintenance planning as well as asset management strategy.

Necessary solution configuration to generate the prognoses is efficient and requires minimal effort: once the critical pieces of equipment have been selected, relevant malfunctions have been defined and suitable condition parameters have been formulated, the configuration is completed and prognostic reports for the equipment at hand can be generated. By using the already existing data history, accurate prognoses can be compiled right from the moment of completion of the configuration. What is more, the precision of the prognoses increases over time due to a machine learning mechanism built into the technology. Applications of the prognostic solution, for example in power stations in the USA and Europe, and in the railway industry, have successfully demonstrated its strength. All in all, it has a considerable and diverse usage potential.

Strengths of the approach

The importance of prognoses for an economic and technical optimization of asset operation becomes clear in practice. Condition-based RUL prognoses for important pieces of equipment are a quantum leap for many maintenance and reliability managers. Alternatives, like the tightening of alarm values in condition monitoring, are neither from a cost nor from a security perspective adequate: such per se measures can lead to an increase in false alarms with a negative impact on various areas of operation.

While many decisions made by asset and maintenance managers relate to the future, they are nevertheless not based on prognostic information. For example, they expect that there will be no incidents before the next routine maintenance. Another example is when companies accept contracts assuming that the machinery needed to complete the job will not fail during the production run. Yet, the decision makers have limited objective assurance when it comes to these future expectations. With Prognostics they gain knowledge about when the time window for malfunctions to occur will open and close again. This information helps to actively control the RUL of the equipment, for example by adjusting the operational load. By knowing when maintenance interventions will become necessary, the maintenance actions can be bundled intelligently -which decreases the total number of costly downtime events. That way maintenance schedules based on fixed intervals can be turned into schedules based on real technical necessity. The result is saved cost and, above all, increased uptime.

Scepticism about working with probabilities

Traditionally, probabilities are not a well-known way of presenting the condition of an asset and its RUL information. Maintenance managers are used to deterministic arguments, i.e. hard facts. Yet, they are often not aware that they are actually implicitly working with probabilities, for example when they base their future maintenance plans on past experience or manufacturer’s recommendations. A prognostic solution explicitly calculates such probabilities in a comprehensible and transparent manner, free from human mistakes. This provides the maintenance manager with a very valuable instrument to make better decisions or, for example, to prioritize maintenance interventions intelligently.

As an additional benefit, a prognostic report delivers diagnostic information on the current condition of the components. Apart from the prognostic and diagnostic strengths, a prognostics-based solution helps with consolidation, prioritization and management of condition data. For many companies, such as operators of power plants and industrial facilities, the prognostic solution is the only medium that can consolidate data and findings, for example from vibration and lubricant analyses.

The future lies in Prognostics

Both condition Diagnostics and condition Prognostics profit from data collection innovations, for example in the context of the Internet of Things, as hardware innovations have led to robust, wireless vibration and corrosion sensors. Software innovations in data management enable the use of large and remote data sources..

But even without making full use of these recent developments, many operators’ data histories are sufficient in order to extract value from Prognostics. On top of this, the transparency provided by the solution allows to add or, in certain cases, even eliminate data sources in a targeted manner based on actual demand, instead of a cost intensive expansion of the overall database.

Even though great progress is being achieved in Diagnostics, Prognostics is essential for condition-based maintenance: machine operators need RUL prognoses to optimize maintenance plans for the early detection of malfunctions and for the avoidance of downtime. Service providers from the condition monitoring sector are interested in automated intelligent prognoses, for example to scale and differentiate their services. Software providers from the maintenance planning and asset management sectors want to offer prognostic models in order to create more degrees of freedom for maintenance cost optimisation.

The above points show that the aims of Prognostics and automation are woven together. Along with a predictive maintenance philosophy and the knowledge around relevant malfunctions and suitable condition parameters, a complete and detailed condition and process data history becomes increasingly important. The RUL prognoses including all ancillary functions provide a solid foundation to stay abreast of the increasing pace of technological developments. Maintenance managers receive a tool to continuously improve and validate their decisions.

Moritz von Plate has been CEO of Cassantec (, a company specializing in Prognostics for industrial assets, since 2013. Prior to joining Cassantec he was CFO at Solarlite, an EPC Contractor for concentrated solar power plants. Moritz von Plate also worked seven years at the Boston Consulting Group, focusing on industrial goods and financial services industries. He has an MBA from Georgetown University and is an agricultural engineer.

Disclaimer: The views and opinions expressed in this article are those of the author and do not necessarily reflect the opinion or position of Service in Industry or its owners