Method for contactless diagnosing power facility using artificial intelligence and signal processing technology and device using the same

A power facility diagnostic device using artificial intelligence, comprising:

• a communication unit;
• a display;
• and
• a control unit,
• wherein the control unit is configured to:
• acquire a plurality of ultrasonic signals of the power facility through the communication unit;
• generate a plurality of mel spectrograms based on the plurality of ultrasonic signals;
• create a plurality of training data sets by labeling each of the plurality of mel spectrograms using an annotation tool by:
• displaying, simultaneously via a user interface of the annotation tool on the display, a waveform of an ultrasonic signal to be labeled and a mel spectrogram corresponding to the ultrasonic signal to be labeled,
• receiving, via the user interface, a selection of a type for a labeling item, wherein the type is selected from a group consisting of an arc state, a tracking state, a corona state, and a normal state,
• receiving, via the user interface, a setting of a position and a length for the labeling item on the waveform or the mel spectrogram,
• separating a portion of the ultrasonic signal to be labeled according to the set position and length, and
• independently storing original data comprising the separated portion of the ultrasonic signal, refined data comprising a waveform or a mel spectrogram corresponding to the separated portion, and labeling data in which the labeling item is added to the original data and the refined data;
• train a diagnostic model based on the plurality of training data sets, wherein the diagnostic model is based on a convolution neural network algorithm;
• input a new mel spectrogram generated from a new ultrasonic signal into the trained diagnostic model;
• and
• acquire diagnostic data for the power facility based on an output of the trained diagnostic model.