Big data and intelligent maintenance

Data-based prognostic technology can determine the future condition of machines, laying the foundation for intelligent maintenance planning, writes Moritz von Plate

The world’s energy needs are constantly growing. Worldwide population growth and the continuing industrialisation of emerging economies, notably China and India, are the major causes for this growth in energy consumption, which has a negative impact on the environment. According to the Intergovernmental Panel on Climate Change (IPCC), anthropogenic greenhouse gas emissions, i.e., emissions caused by human activity, have increased significantly since pre-industrial times and are currently at an all-time high. Green technologies, such as cogeneration plants, have therefore become increasingly relevant for energy production and will become even more relevant in the future.

Thanks to the new technologies of the Internet of Things, it is now possible to perform cost-effective maintenance measures that can increase security and prevent unplanned outages in cogeneration plants. Such new technologies make it possible to analyse process and condition data of plants and make prognoses of the system’s future state. In addition, these prognoses change the way in which people make decisions.

The role of data

The industry is offered totally new possibilities through the Internet of Things, especially when it comes to process optimisation and automation. The way has been paved for profound changes to industrial processes by implementing modern information technologies. In the course of advanced digitalisation, machines are linked with one another and collected data is used to intelligently co-ordinate and improve processes. When it comes to maintenance and operational management, Big Data technologies enable a data-based and future-oriented prognostic strategy.

For example, thanks to innovative Big Data technologies, prognoses on the future condition of a machine or its individual components can be created. With a prognostic approach, users receive a data-based prognosis and can adjust maintenance plans accordingly. Further, unnecessary costs or unplanned outages can be avoided, for example by replacing parts in time, i.e., not too early and not too late. In this context, prognostics can be defined as an ‘objective and data-based forecast of future conditions with an explicit time reference’. In practical terms, this means that prognostic reports can provide information on the future condition of machines or machine components for a period of mostly weeks or months or, in special cases, even years.

Predictive diagnostics vs prognostics

This prognostic approach is not synonymous with the so-called Predictive Diagnostics or Predictive Analytics. Predictive Diagnostics recognises initial early warning indicators for future malfunctions by means of data abnormalities, and provides diagnostic findings about the current condition. Yet it does not provide information on when an abnormality will turn into a malfunction, i.e., when the time frame until the next malfunction arises will close (tomorrow, in a week, or is it still months?). Prognostics, on the other hand, not only reports on when one can expect a malfunction, but also indicates when the time frame during which measures can be taken will close.

Because the prognoses are calculated for each machine individually, they are not based on average data from other machines or manufacturers’ specifications. This has the advantage that the individual performance curves, the operational strategy and, if applicable, previous data on historical incidents is included in the prognoses. This results in the prognoses reaching a higher level of precision and reliability. When calculating prognoses, the historical data runs through a number of different steps. These consist of stochastic methods and include highly developed algorithms. The result is an explicit future risk profile that illustrates the probability of malfunctions over time.

An illustrative excerpt from a prognostic report for one example generator

Source: Cassantec


The colour green represents a low risk of malfunction

Source: Cassantec

The requirement for a prognosis is to collect and store enough process data (e.g., rotation frequency, speed, temperature and pressure) and condition data (e.g., vibration data, lubrication data and housing temperature). An ideal time frame of data history is three to five years, whereby it is possible to complete a reliable prognosis with a shorter time-frame. The storage format does not play an important role. It is more important to ensure that the data is as complete as it can be, as this will increase the validity of the statistics.

Condition-based maintenance

Instead of relying on fixed maintenance intervals or waiting for something to break, the information from a prognostic report can be used to ensure that maintenance and repair work can be carried out when needed. Parts will not be replaced too early on speculation, but rather when it is necessary from a technical point of view. Apart from this, by means of the prognostic reports and good data processing, it is also possible to recognise the effect that various operational scenarios will have on the equipment’s remaining useful life (RUL), transparently and objectively. By doing so, the RUL can be actively managed through adjusting the operational mode.

How the installation works

Introducing transparency into the RUL and, ideally, being able to actively control it were the aims of a project in which Cassantec implemented the solution in a fossil fuel-fired power plant. The active management of the RUL should take place in such a way that the duration of the RUL and the operational mode are balanced to achieve the desired outcome. Additionally, maintenance activities should be optimised to lower the operational and repair costs.

Such a project is divided into two phases. As a prerequisite, historical available condition and process data from the power plant must be collected and prepared for further processing. During the first phase – the so-called configuration phase – the power plant experts and Cassantec ascertain the correlations between data parameters and specific malfunctions. The second phase is prepared based on this foundation: the actual calculation and prognoses of the risk of malfunctions. This phase also includes the fine-tuning of the preliminary component specific warning and alarm levels.

How the solution works at a cogeneration plant

The first prognostic reports compiled for a cogeneration plant have already delivered valuable findings for the operator. For example, by implementing a scenario analysis which determines the dependence of the data on the operational regime, it is possible to find a new and optimised mode of operation for the equipment. This can have a positive effect on the RUL of the equipment, its reliability and the need for maintenance.

Based on results produced by the prognostic solution, the energy provider receives valuable insight into the relationship between operational strategy and the RUL of the power plant and, in particular, the critical equipment. This goes much further than the information available from conventional condition monitoring and diagnosis. The results enable the operator to make well-founded decisions on the adjustment of his or her operation and maintenance plan for the critical equipment, in order to be able to optimise its usage in three fundamental aspects: considerable extension of the RUL, minimising maintenance costs through optimisation of the maintenance plan, and specific information on when a component will need to be replaced.

When the operator decides to expand the implementation of the prognostic solution to other similar plants in the fleet, the configuration phase, as outlined above, is significantly shortened. In addition, the operator can expect extensive savings in maintenance and repairs, and a comprehensive understanding of the condition of the machinery and of the factors that influence the RUL. Fleet-wide implementation also leads to a fleet-wide learning effect that boosts the initial advantages.

How people will make decisions in the future

The dots show the exact data reading points

Source: Cassantec

Whether consciously or unconsciously, humans make hundreds of choices every day. Gerhard Roth, a professor at the Institute for Brain Research in Bremen, has determined that, quite often, gut decisions are the better choice. When choosing what to eat for breakfast or what to wear, that is perhaps the best way; however, for more complex decisions the basis should not be intuitive. Especially when the cause and effect of a problem are not clear and decision-makers are faced with complex structures, data-based facts can put them on the right track. Algorithms help people solve complex problems such as the maintenance of equipment, and help them make better judgments.

At present, the basis for making many decisions is still often experience or intuition. Humans have their own ‘computer’, the brain. However, the brain is not immune to prejudice. Even factors such as the weather or one’s mood demonstrably and significantly influence decisions. Often many important characteristics are lacking for a proper analysis and assessment, but an algorithm that is programmed in advance is subject to fewer such errors in reasoning. Mathematical foundations offer the possibility that decision-makers receive a formula that is objective, transparent and applicable to different situations.

Thus, for example, through the use of Cassantec’s prognostic reports, a foundation is created to make sound decisions for maintenance strategies – for example, to pool maintenance interventions intelligently and to plan them in time to avoid costly overtime and night shifts. Maintenance plans will no longer be created periodically and based on experience, but with a transparent, data-based structure. This saves companies huge costs.

What is holding us back

Society is at the beginning of a digital transformation. Industry 4.0 and the Internet of Things offer enormous potential to change and exercise a positive influence over the way employees work. Yet technologies such as prognostics also face challenges. The prudent application of prognostic solutions requires that reliability and maintenance professionals possess an extended skillset: the ability to articulate risk, to explicate forecasts, and to consider both in asset management decisions. Prognostics complements and requires operator experience and manufacturer know-how, but it also necessitates a shift in thinking and language towards a risk management approach. In the long run, though, it is clear that companies and professionals must face these challenges. Companies that have not already started collecting data for sophisticated analyses, and that are not planning to make use of the new possibilities, will eventually reach the point where they can no longer compete in the digitalised environment.

The foundation for intelligent planning

The use of complex data analytics in order to control and improve processes is increasing in the age of Big Data and the Internet of Things. When it comes to maintenance and repair activities, the use of big data analytics is likewise increasing. With the help of data-based prognostic technology, the future condition of machines can be determined. This creates the foundation for intelligent maintenance planning. Instead of fixed intervals, maintenance will now only take place when it is technically necessary. Implementation in a cogeneration plant can increase the understanding and transparency for the plant. The foresight derived from prognostics can enable an active control and expansion of the RUL.

Advantages of prognostics:

ࢀ¢ Maintenance can be carried out when it is technically necessary, which reduces the number of maintenance interventions;

ࢀ¢ The influence of the operational regime on the RUL becomes transparent, which means that it is possible to actively manage RUL;

ࢀ¢ It becomes apparent well in advance when the risk of a malfunction will reach the risk tolerance threshold. This allows for avoidance of unplanned malfunctions;

ࢀ¢ Repairs can be planned in advance and then conducted when the impact of operational interruptions is at its lowest;

ࢀ¢ The processing and presentation of the data provides transparency and enables fleet-wide comparisons over time;

ࢀ¢ Decision-making competency can be increased by means of objective information, the machine will gain in safety and reliability, and the reduction of (unplanned) malfunctions will save budget.

Moritz von Plate is CEO of Cassantec

This article is available on-line. Please visit

No posts to display