Methods of analyzing cement integrity in annuli of a multiple-cased well using machine learning

A method of characterizing the status of annuli of a multiple-cased well of interest, comprising:

• utilizing a dataset of labeled slowness time coherence (STC) samples obtained from processing at least one of (i) synthesized sonic data and (ii) measured sonic data obtained from a multiple-cased well, training a machine learning module to receive STC sample inputs and provide output labels for a plurality of states for annuli of a multiple-cased well;
• locating a sonic tool having at least one transmitter and multiple detectors at a location in the well of interest;
• firing the at least one transmitter, and detecting with the multiple detectors the resulting sonic waveforms impacted by annuli of the well of interest;
• preprocessing the sonic waveforms to obtain at least one STC map, wherein said preprocessing the sonic waveforms comprises bandpass filtering said sonic waveforms into at least two bands and conducting STC processing on each band separately;
• providing the at least one STC map as the STC sample inputs of the machine learning module to obtain an indication of the status of the annuli of the multiple-cased well adjacent the location of the sonic tool.