Predictive maintenance: What is it and steps to do it correctly?

Machine and equipment failures could cost companies large sums of money, in the form of production stoppages and cost overruns for urgent interventions: ensuring their proper functioning becomes an almost mandatory task to react to unforeseen situations. We are going to analyze what predictive maintenance is all about, with examples, techniques and tools to illustrate this new approach to anticipation.

What is predictive maintenance

Predictive maintenance consists of a series of techniques and tools that, based on the analysis of data, manages to detect anomalies and possible errors in the operation of processes and equipment.

Through a detailed analysis of real information from equipment and procedures, this maintenance system is able to detect trends and patterns of operation: the result is a scenario in which errors are easily predictable, because concrete data is being used to make the prediction.

As can be easily inferred, predictive maintenance is based on anticipation and forecasting, but not by sticking to potential errors arising from elucubrations or a priori hypotheses, but only relying on real data, collected for that purpose.

In this way, predictive maintenance can anticipate possible failures monitoring the condition of the machinery, without speculations or theories, forming a constituent part of the IIoT (Industrial Internet of Things).

Predictive maintenance techniques

Some predictive maintenance techniques widely used in industry include:

• Vibration analysis and detection, to analyze rotating machinery.
• Infrared thermography to inspect insulation, steam traps or electric motor stator failures.
• Analysis of lubricants, to detect deterioration due to contaminants and wear particles.
• Partial discharges in rotating electrical machines in operation cycle.
• Ultrasonic analysis, to verify sealing parameters, valve operation or fluid leak detection.

Each of these has several practical applications, which we’ll look at in the following examples.

Examples of predictive maintenance

Predictive maintenance has a multitude of applications in industrial environments: for example, to reduce production stoppages due to defective parts entering the assembly line, or to analyze vibrations that may precede possible misalignments, looseness, unbalance or wear.

It is also possible to identify if industrial equipment will start to malfunction from very early detection of signs, such as changes in temperature; it is also possible to detect leaks in tanks and piping, mechanical errors in moving parts and failures in electrical equipment.

Even the classic vending machines are susceptible to receive this type of maintenance, going a step beyond the traditional periodic technical reviews: it is possible to improve the lifetime of these machines and reduce the number of repairs that they would have to undergo if they had been subjected to otherwise.

Do you need to manage documentation, work parts, invoicing, warranties and reviews of your assets installed at your customers? Take your business to the next level with the best-rated SAT software on the market.

Steps to perform predictive maintenance

1. Monitoring machine operation

In this first stage, the central part consists of having digitized data on the status of the machinery, and that such reading is done automatically and permanently, i.e., without human intervention.

It is evident that each process or equipment will have different relevant parameters, as well as different reading frequencies: it is not the same to measure, for example, the position parameters in a pressing process than in one where a robotic arm intervenes.

2. Process modeling and directed maintenance

When you have control over the reading of the relevant parameters of the equipment and processes, a stage begins consisting of generating a model based on historical data, where you can see what behavioral trajectories have been followed, and under what circumstances.

Algorithms are designed at this point to relate the parameters to each other, in order to detect patterns that are repeated or that always occur under the same environmental conditions, or in the face of the cross-influence between them. As more historical data becomes available, system learning will become more complete, allowing more accurate conclusions to be drawn.

Therefore, at this point it is possible to start designing a normal behavior model of the equipment, that is, how the assembly is supposed to respond under usual environmental conditions: once this pattern has been assumed, it will be possible to prioritize predictive maintenance on those areas that present a greater deviation from the normal operating curve (targeted preventive maintenance).

3. Modeling limit scenarios

Limit scenarios are understood as those operating environments under whose conditions it will be more likely that the machinery will fail and will not be able to continue working: with this definition, it is possible to close the circle and delimit the operating scenarios, since both the “normal curve” and its dimensions are available.

Once again, the more information gathered from these limit scenarios, the greater the probability of detecting the situations in which the machinery will stop working; consequently, a functional model can be drawn with greater accuracy.

4. Predictive maintenance

Once you reach this stage, you will have a much more accurate view of the maintenance work that will be necessary to deploy on certain machines, since you will have been able to refine the probability with which a piece of equipment will not work, and under what conditions it will fail.

In this way, it will be possible to prioritize the truly necessary or critical actions over the accessory or less urgent ones: this filtering of interventions allows the company to improve the coordination between the production and planning departments.

5. Continuous monitoring

The fifth stage of predictive maintenance is simply to assume its cyclical nature and to carry out monitoring to further improve data collection systems and predictive behavioral models.

Only thanks to this feedback, predictive maintenance systems can learn on the fly and be more accurate in identifying patterns: monitoring tasks require, necessarily, the definition of some key performance indicators (KPI) of the system, which provide sensitive information about their level of accuracy and probabilistic accuracy.

Difference with preventive maintenance

Predictive maintenance, as we have seen, is a fully proactive method, based on real-time data collection from the equipment: this allows the technical managers to make a prediction of the behavior of the machinery, including, of course, possible breakdowns.

However, the preventive maintenance consists of periodically checking the equipment so that it works properly, and thus avoiding equipment failures before possible incidents occur. It is based on a series of checks suggested by the manufacturer, and whose manual review every so often (or as soon as a typical “symptom” is detected) guarantees, on paper, its correct operation.

Predictive maintenance is postulated as one of the solutions that best leverages data science and the Internet of Things to optimize the work of the TSS. Through a series of specific tools and techniques, it is possible to model machine behavior scenarios and predict possible errors, allowing the company to better distribute and coordinate its service resources.