Goal

Determine the equipment operation modes and anomalies with the use of data from the vibration sensor.

Our solution

We constructed a shaker with two electric motors powering the vibrations in it. We used the shaker to simulate the work of the machine.

We developed an autoencoder based on the neural network on the PyTorch engine. Our solution receives and restores vibration signals, comparing the restored result with the original one. Thus, we receive information about deviations in the equipment operation.

The autoencoder helped us to determine 2 operating modes of the shaker: work with one and with two powered motors.

Next, we plotted the deviations in the work of the shaker on a graph.

A rod was attached to the axis of one of the motors. The rod simulated a connection subject to wear and leading to malfunction.

The graph below clearly distinguishes the moment of attachment of the rod to the electric motor axis (from the 120 mark to the 148 mark).

Right after it the operating mode anomaly starts progressing - and at the mark 175 conditional connection break takes place.

Thus, using a neural network trained in the normal operation mode of the equipment, we can both distinguish the normal operation modes of the equipment and accurately predict the anomalies emergence.