Predictive Maintenance

Predictive Maintenance: More Efficient Maintenance with Artificial Intelligence

Predictive maintenance means “predictive maintenance/servicing”. This means that for each wear part of a device or a technical system, it is continuously estimated when the next maintenance is necessary. This estimation is based on the actual condition of the component – rather than carrying out maintenance when a certain number of operating hours or mileage has been reached.

This would mean, for example, that the engine oil of a vehicle is not always replaced exactly when 10,000 km is reached, but that various quality parameters of the oil (e.g. concentration of various foreign substances) are continuously measured and how long it will take before the oil can no longer be used safely.

Now think of any piece of equipment or technical equipment that needs to be maintained from time to time, such as a ventilation system. Which components need regular maintenance? Ball bearings, gears, drive belts, filters, … In many cases, it is possible for an experienced employee, who has carried out exactly this maintenance many times before, to estimate quite well from the noise, vibrations, heat generation or other variables in which condition a wear part is at the moment, if he looks at the thing closely enough.

An AI system for predictive maintenance works in a very similar way. First of all, those measured variables must be identified that provide good indications of the condition of a component and are sufficiently easy to measure.

Which measured values are particularly suitable

  • Vibrations: These are essentially the same as the sounds that a device makes, making them a very meaningful measure of the condition.
  • Current consumption: in particular, the starting current (i.e. the current consumed by an electric motor in the first few seconds after being switched on) can also provide valuable information
  • Temperature: air temperature, temperature of individual components/parts of a device, temperature of liquids, (friction!)
  • Flow of gases or liquids through pipes
  • Pressures, differential pressures, …

Early indications of an imminent need for maintenance can sometimes only be detected by looking at different measurands at the same time.

Then the course of those sizes that have been selected as suitable must be recorded over the entire life cycle of the wear part. This can also be used to train a neural network that independently identifies early indications, learns the typical course of the measured variables over the life cycle of the component and can estimate when the next maintenance will be necessary.

In other words, the neural network can do things that are similar to the skills of an experienced employee. But in contrast to a human employee, the AI works tirelessly day and night, incurs only low costs in operation, and – if the sensors are correctly attached – has finer sensory organs, so to speak: It can perceive vibrations, noises, temperature changes, and much more that remain hidden from a human.

How you can benefit

  • Avoidance of unplanned downtime: Unplanned downtime can be incredibly costly, for example if it affects an entire production line or even an entire factory. So it’s obvious to what extent predictive maintenance can help.
  • Fewer routine checks: When there is solid data about the health of a device, it is possible to perform routine checks and checks far less frequently. This saves working time.
  • Avoidance of unnecessarily early maintenance work: If it is known in what condition a wear part is located, it is not necessary to replace it far too early as a precautionary measure in order to prevent damage in any case. Instead, such a component can be used optimally until the end of its service life. In addition to the cost savings in materials, this of course means a reduction in the necessary working time.

Predictive Maintenance vs. Anomaly Detection

Now we want to differentiate predictive maintenance from another important method of applied AI, namely anomaly detection. Predictive maintenance estimates how long a device can continue to operate, while anomaly detection only detects that there is some change that may indicate damage.

Want to know more?

Do you have any questions about artificial intelligence in the Internet of Things? Or an exciting use case? We look forward to hearing from you.

Similar Posts