Abnormal sound detection device and abnormal sound detection method

An abnormal sound detection method for inputting an operating sound of machine equipment to be diagnosed by a sound collection device and determining presence or absence of an abnormal sound by a calculating machine, the calculating machine having

• a process of inputting a normal operating sound of the machine equipment during a first learning process, decomposing a normal sound base spectrum from an amplitude spectrogram, and storing the normal sound base spectrum in a first database,
• a process of extracting a first new sound/new vibration component from the normal operating sound of the machine equipment during a second learning process, learning a post-new sound/new vibration extraction normal sound model from the first new sound/new vibration component, and storing the post-new sound/new vibration extraction normal sound model in a second database,
• a process of executing supervised nonnegative matrix factorization (NMF) using the post-new sound/new vibration extraction normal sound base spectrum of the first database as teacher data on an acoustic signal of the machine equipment input during diagnosis, and outputting a nonnegative matrix not allowed to be approximated with a low rank in the normal sound base spectrum as a second new sound/new vibration component, and
• a process of computing a probability that the second new sound/new vibration component extracted from the acoustic signal of the machine equipment during diagnosis is generated based on the post-new sound/new vibration extraction normal sound model previously stored in the second database and determining whether or not the second new sound/new vibration component is abnormal.

An abnormal sound detection device for determining presence or absence of an abnormal sound using an operating sound of machine equipment to be diagnosed as an input, the abnormal sound detection device comprising:

• a non-transitory computer readable storage medium storing at least one program;
• a processor configured to execute instructions contained in the at least one program to execute the steps of:
• inputting a normal operating sound of the machine equipment during a first learning process, decomposing a normal sound base spectrum from an amplitude spectrogram, and storing the normal sound base spectrum in a first database,
• extracting a first new sound/new vibration component from the normal operating sound of the machine equipment during a second learning process, learning a post-new sound/new vibration extraction normal sound model from the first new sound/new vibration component, and storing the post-new sound/new vibration extraction normal sound model in a second database,
• executing supervised nonnegative matrix factorization (NMF) using the post-new sound/new vibration extraction normal sound base spectrum of the first database as teacher data on an acoustic signal of the machine equipment input during diagnosis, and outputting a nonnegative matrix not allowed to be approximated with a low rank in the normal sound base spectrum as a second new sound/new vibration component, and
• computing a probability that the second new sound/new vibration component extracted from the acoustic signal of the machine equipment during diagnosis is generated based on the post-new sound/new vibration extraction normal sound model previously stored in the second database and determining whether or not the second new sound/new vibration component is abnormal.