Event detection from pump data

A system, comprising:

• at least one processor;
• memory in electronic communication with the at least one processor;
• and
• instructions stored in the memory, the instructions being executable by the at least one processor to:
• receive historical pressure data collected with one or more pressure sensors at a plurality of wellbores, the historical pressure data comprising a plurality of historical pressure pulses within drilling fluid at the plurality of wellbores, the plurality of historical pressure pulses being associated with a plurality of downhole operations of a plurality of downhole tools within the plurality of wellbores;
• generate a plurality of historical pressure spectrograms from the historical pressure data, the plurality of historical pressure spectrograms each representing a spectrum of frequencies of the plurality of historical pressure pulses over time;
• identify, independent of the plurality of historical pressure spectrograms, a plurality of historical downhole events that occurred during the plurality of downhole operations;
• correlate the plurality of historical downhole events to the plurality of historical pressure spectrograms to generate labelled training pressure spectrograms;
• train a pressure spectrogram machine learning model based on the labelled training pressure spectrograms to identify pressure signatures in the labelled training pressure spectrograms corresponding to the identified plurality of historical downhole events, the pressure spectrogram machine learning model being trained to identify downhole events from input pressure spectrograms;
• identify, with a pressure sensor at a surface of a target wellbore, target pressure data comprising a plurality of target pressure pulses within a flow of a target drilling fluid, the plurality of target pressure pulses being associated with a target downhole operation of a target downhole tool;
• based on the target pressure data, generate a target pressure spectrogram;
• using the target pressure spectrogram as input to the pressure spectrogram machine learning model, identify a target downhole event affecting the target downhole operation;
• and
• based on identifying the target downhole event, generate a notification indicating an abnormal operation of the target downhole tool.

A computer-readable storage medium including instruction that, when executed by at least one processor, cause the processor to:

• receive historical pressure data collected with one or more pressure sensors at a plurality of wellbores, the historical pressure data comprising a plurality of historical pressure pulses within drilling fluid at the plurality of wellbores, the plurality of historical pressure pulses being associated with a plurality of downhole operations of a plurality of downhole tools within the plurality of wellbores;
• generate a plurality of historical pressure spectrograms from the historical pressure data, the plurality of historical pressure spectrograms each representing a spectrum of frequencies of the plurality of historical pressure pulses over time;
• identify, independent of the plurality of historical pressure spectrograms, a plurality of historical downhole events that occurred during the plurality of downhole operations;
• correlate the plurality of historical downhole events to the plurality of historical pressure spectrograms to generate labelled training pressure spectrograms;
• train a pressure spectrogram machine learning model based on the labelled training pressure spectrograms to identify pressure signatures in the labelled training pressure spectrograms corresponding to the identified plurality of historical downhole events, the pressure spectrogram machine learning model being trained to identify downhole events from input pressure spectrograms;
• identify, with a pressure sensor at a surface of a target wellbore, target pressure data comprising a plurality of target pressure pulses within a flow of a target drilling fluid, the plurality of target pressure pulses being associated with a target downhole operation of a target downhole tool;
• based on the target pressure data, generate a target pressure spectrogram;
• using the target pressure spectrogram as input to the pressure spectrogram machine learning model, identify a target downhole event affecting the target downhole operation;
• and
• based on identifying the target downhole event, generate a notification indicating an abnormal operation of the target downhole tool.