Predictive maintenance is a methodology that applies to different industries, but all of which share the need to make their processes smoother and at the same time less onerous: in a word, more efficient.
In order to understand what it is, we first need to highlight the fundamental elements and steps that compose it.
Anomaly Identification, observing the present to get a prediction: examples and methodologies
The first and essential phase for achieving predictive maintenance is the observation of real, present phenomena as they occur.
Anomaly Identification is a good starting point because it allows companies to recognize anomalies that have already taken place or are happening while the analysis is being carried out, in the most punctual and precise way they can be identified.
For example, for the automotive sector, in the very specific case of car painting, quality control is carried out through various vision systems – such as cameras, or directly through the operator’s human eye – which are often unable to detect all anomalies in real time.
A data-driven system, with computer vision technology, capable of combining optical inspection with 3D scanning to analyze images or spectrometry relating to each car, helps to identify any anomalies faster and more accurately.
When you can analyze a phenomenon so precisely on a numerical level, you have a clearer perspective on how, by changing one specific element, all the others change accordingly. This is the first step towards a real prediction of the anomaly.
Nowcasting, predicting the present: an example in the Gas & Oil industry
In this sense, one of the most effective methods of Anomaly Identification is the so-called Nowcasting: prediction of the present.
In practice, a phenomenon is observed with all its variables related to each other, and eventually the system establishes an optimal scenario according to which certain values will be obtained from the other variables’ variation.
In this way it is possible to predict the behavior of the variables that cannot be monitored.
As a matter of fact, the system compares the real values - obtained by monitoring the process with instruments dedicated to detection – with the expected values, i.e. those that should result according to the forecast of that specific process at a given moment.
When these two values deviate beyond a certain threshold, they must be interpreted as an anomaly, an alarm signal. There is therefore a problem that could not have been detected without this approach, and of which we are now aware.
Considering that a frequent occurrence of these deviations heralds a malfunction, a probable failure, it is precisely the sum of these variables that provides companies with important information on what exactly may be compromising the system.
For example, in the field of gas detection there are electronic and connected meters that detect how much gas passes from the network to the private domestic heating system. Based on this information, the meter operator informs the gas sales company how much gas has actually been consumed, in order to issue the invoice.
But what happens when the meters go haywire? In this case the figure sent to the gas sales company will produce a much higher invoice than the actual consumption: this means a very significant cost at the end of the year.
Speaking of which, how does the gas sales company carry out an assessment of broken meters and schedule inspection visits to proceed with replacements by prioritizing the most urgent interventions?
Having so many delivery points makes a simultaneous manual analysis difficult, and this is why analyses are based on the behavior and consumption of certain classes of users divided by geographical area, family class, type of property, combining those information with exogenous data (such as the weather, etc).
In this way, consumers are clustered in a specific reference class, in order to identify a hypothetical anomaly through the prediction that emerges.
Anomaly Identification and Predictive Maintenance in Wind Energy
In fact, Anomaly Identification is a precise picture of a given situation, where the anomaly is promptly discovered by comparing the data obtained from the machinery and those of nowcasting, checking if these two quantities deviate even slightly.
Another example of predictive maintenance comes from the wind energy industry, where a different future-wise forecasting is applied..
In this case, there is a system that tracks malfunctions in wind generation: a history of failures based on all the episodes involving the various elements of the turbine, the default episodes of the inverters and strings of the hydraulic system, etc.
All the information on what is happening inside the machinery, collected numerically by the CMS and data loggers, is related to each other, comparing the behavior of the machine with the specific fault in question. This helps to identify patterns, i.e. to understand which repeated “action and reaction” lead to a malfunction, in order to be able to prevent it in time.
Therefore, this is a future forecast based on the observation of historical data sets, which basically only describes “what could happen inside the machinery”.
To understand which actions you should take, a further step is needed: prescriptive maintenance.
From Predictive to Prescriptive: how to enhance Maintenance
To carry out this type of maintenance, you can proceed in two different ways, depending on the case.
When certain detected phenomena signal the possibility of failure of a specific component within a certain period of time, the system can prescribe:
- an extraordinary maintenance intervention or the component replacement
- a targeted ordinary maintenance plan
This is all about efficiency: the point of balance between business continuity and profitability.
In case of over-maintenance of an asset or continuous replacement of its components, the piece obviously would not break. But this approach costs money, and if you run a business you know there are always two main objectives: to keep the production smooth, but also to avoid waste and contain costs.
What should you do, then?
Process efficiency is achieved when what is spent on maintenance and replacement of a component is less than what would be lost due to its downtime and repair costs.
Prediction and prescription: an example in the Solar Energy industry
To keep the maximum efficiency of photovoltaic systems, there are three possible scenarios to consider, with three respective malfunctions that can inhibit energy production:
- inverter breakdown;
- panel breakage;
- panel that gets dirty (the cells are obscured by dust/mud etc) and therefore captures less radiation
Prendiamo come esempio il terzo scenario: ogni quanto occorre pulire i pannelli?
Una possibile strategia è quella di monitorare costantemente, attraverso dei controlli periodici, e nel momento in cui si rileva più sporco del dovuto, si chiama la società di pulizie o si affida il lavoro al tecnico in azienda.
Let’s take the third scenario as an example: how often should the panels be cleaned? A possible strategy is to constantly monitor, through periodic checks, and when too much dirt is detected, you call the entrusted technician or the cleaning company to fix it.
But the fact is that you cannot always access these readings directly, and therefore you should obtain these specific information from the system, which is able to highlight when a drop in performance is potentially due to dirt on the panel.
The next step is to predict what could happen in the future. This way you can schedule cleaning interventions according to the most relevant factors, such as meteorology, which has a significant influence.
In this sense, prescriptive maintenance can be done in two ways:
- providing a different and more targeted rule for scheduling interventions;
- through predictions that triggers specific extraordinary activities to improve system performance.
In short, through the description of a phenomenon it is possible to make predictions, which are the basis of subsequent prescriptions, in a flow that considers all the variables mentioned.
In this sense, risk analysis is part of descriptive analysis: the most correct assessment for the so-called relevant risks of an asset (we are talking about environmental risks, health risks such as worker safety).
The basic objective is to optimize the cost/opportunity ratio of carrying out maintenance on the machinery in time, through a forecast of possible malfunctions, rather than replacing all the parts more frequently.
In fact, there are many circumstances, often linked to budget constraints, in which it is not possible to make unclear assessments of interventions, but it is necessary to make more specific forecasts to prescribe extraordinary maintenance.
In this way it is possible to reach more informed decisions, improving industrial processes and making them more efficient.