System based on neuron network for detection and prediction of failures of monitored devices. The system assesses device expression (sound, vibrations, temperature, pressure, stress, etc.) and assesses them, via neuron networks, in a way similar to how they would be assessed by an experienced user (servicer) of the device. The system can detect device flaws (defects) also several months in advance and thus avoid unplanned system downtime. Thanks to timely repair, this leads not only to cost savings for repairs themselves and on unplanned downtime, but also protects the health of persons servicing or using the devices.