Russian Scientists Train AI to “Hear” Faults in Electric Motors
Neural network learns from synthetic yet realistic failure patterns
Researchers at the Institute of Artificial Intelligence and Digital Sciences at the Faculty of Computer Science at HSE University have developed a method called Signature-Guided Data Augmentation to detect faults in industrial electric motors.
Until recently, engineers identified faults by analyzing the electric current signal consumed by a motor. Specific frequencies were extracted from the signal, and different parameters were checked manually. However, this process is labor-intensive. Training machine learning models has also been challenging due to a lack of data on how motors behave during failures.
The HSE research team developed a way to replace manual analysis with fast automated diagnostics. Their approach enables an algorithm to generate synthetic faults within the signal of a functioning motor by injecting specific frequencies. This allows the neural network to learn how to detect defects automatically.
For All Types of Motors
The method was tested on data from two motors. Fault detection accuracy reached 99%, while classification accuracy for different types of failures reached 86%. The team now plans to test the tool in industrial settings.
This approach is particularly useful for companies that lack historical failure data or experience with equipment breakdowns. It can also be applied to motors with a wide range of specifications.
The development could help reduce maintenance costs, minimize downtime, and improve industrial safety.
