SENSOR-INDEPENDENT MACHINE FAULT IDENTIFICATION Russian patent published in 2023 - IPC G16Z99/00 G01M15/00 G06N20/20 

FIELD: monitoring machine tools.

SUBSTANCE: invention is related in particular to identification of faults in machine tools by monitoring them. A method for identifying a malfunction of at least one machine tool includes configuring the first plurality of sensors connected to the corresponding first plurality of machine tools to receive the first plurality of signal sets emanating from the first plurality of machine tools, the first plurality of machine tools having at least one common characteristic, supplying at least the first plurality of sets of signals of the first plurality of machine tools to an existing fault classifier, previously trained with the possibility of automatically identifying faults of the second plurality of machine tools based on signals emanating from them and previously received by the second plurality of sensors, the second plurality of sensors is of a different type than the first plurality of sensors, the second plurality of machine tools has at least one said common characteristic, modification of the specified existing fault classifier by using transfer learning based on at least the first plurality of signal sets of the first plurality of machine tools, thereby providing a modified fault classifier, applying the modified fault classifier to at least one additional set of signals received by at least one sensor from the first plurality of sensors and coming from at least one given machine tool having at least one said common characteristic, wherein the modified fault classifier is configured to automatically identify at least one fault of at least one given machine tool based on at least one said additional set of signals, and provide human readable output data using an output device containing at least an identification of malfunction of at least one said given machine tool, and at least one repair or maintenance operation is performed based on human readable output data.

EFFECT: solving the problem of limited use of a fault classifier trained on data received from sensors of a certain type or types, for classifying signals received from sensors of different types or types, due to differences in the characteristics of the sensors, providing the possibility of transferring learning to match between the original type of sensor, on the basis of which the existing classifier was trained, and new different type of sensors, from which new data are obtained that require classification.

24 cl, 11 dwg