Predictive Maintenance or Why is it broken again?
Machines break down and products break. In the latest report, “The True Cost of Downtime” the company Senseye surveyed 72 international industrial companies about the downtime of their machines. The result: companies lose an average of 323 production hours per year due to machine downtime. The resulting costs for employee idle time, machine restarts and lost sales average $532,000 per hour. It’s a lose-lose situation, because not only are machine breakdowns extremely expensive, they also result in delivery bottlenecks or evendelayed delivery times, all to the annoyance of the company’s own customers. How helpful would it be to be able to predict when a machine or product will break? This idea, which is quite profitable for companies, can be put into practice with the help of artificial intelligence (AI), namely predictive maintenance.
How exactly does predictive maintenance work?
First, an AI uses a machine learning process to learn what signals occur before a machine or product fails. An example: in a sawmill there is a band saw to which a sensor is attached which records the vibrations of the saw. This data is then passed to the AI, which teaches it what the saw’s usual vibration patterns are. In the application, the AI is thus able to give a warning signal as soon as the saw shows unusual vibration patterns. A mechanic can then check it and determine if the saw needs service or if the saw blade needs to be replaced.
The advantage is obvious: each machine can be serviced individually. This saves time and money. An early warning system also allows planning of maintenance and at the same time accidents can be avoided.
But how does the AI behind predictive maintenance work?
In the background, a so-called autoencoder (a special form of neural network) records the saw’s vibrations. The oscillations flow as values through the autoencoder, if it is able to replicate the signal, the signal follows its usual pattern and the saw is fine. However, if the signal cannot be replicated, the signal no longer follows the usual patterns and the autoencoder issues a warning.
In addition to the possibility of using predictive maintenance in production machines, a completely new business field can be opened up for companies. In this way, it is possible to predict the remaining service life of one’s own product and to provide this to the customer as a service.
Do you have your own use case and need support with the implementation or would you like to learn more about the topic of AI? Book our workshop and we will support you in your project.