Abnormality detector of a manufacturing machine using machine learning
An abnormality detector includes a signal output unit for detecting a sign of an abnormality based on a physical quantity acquired from a manufacturing machine and outputting a signal; and a machine learning device including state observation unit for observing, as a state variable representing a present state of the environment, physical quantity data indicating the physical quantity related to an operation of the manufacturing machine from the manufacturing machine; a label data acquisition unit for acquiring, as label data, operation state data indicating an operation state of the manufacturing machine; a learning unit for learning the operation state of the manufacturing machine with respect to the physical quantity, using the state variable and the label data; and an estimation result output unit for estimating the operation state of the manufacturing machine using a learning result by the learning unit and outputting an estimation result.
1. An abnormality detector for a manufacturing machine, the abnormality detector comprising:
a processor configured to:
cause to output a signal in response to a detection of a sign of an abnormality in the manufacturing machine based on a physical quantity acquired from
(1) the manufacturing machine,
(2) a sensor installed in the manufacturing machine or
(3) a sensor installed in a vicinity of the manufacturing machine,
observe, as a state variable representing a present state of an environment, physical quantity data indicating the physical quantity related to an operation of the manufacturing machine from the manufacturing machine, the sensor installed in the manufacturing machine, or the sensor installed in the vicinity of the manufacturing machine,
acquire, as label data, operation state data indicating an operation state of the manufacturing machine,
learn the operation state of the manufacturing machine with respect to the physical quantity related to the operation of the manufacturing machine, using the state variable and the label data, and
estimate the operation state of the manufacturing machine using a learned result, based on the physical quantity related to the operation of the manufacturing machine, and output an estimation result,
wherein, even when a sign of an abnormality in the manufacturing machine has been detected, the processor is configured to still suspend output of the signal, according to the output estimation result of the operation state of the manufacturing machine,
wherein the processor is further configured to
calculate an error between (i) a correlation model for estimating the operation state of the manufacturing machine based on the physical quantity obtained from the manufacturing machine, the sensor installed in the manufacturing machine, or the sensor installed in the vicinity of the manufacturing machine, and (ii) a correlation feature identified from teacher data, which is obtained from the physical quantity obtained from the manufacturing machine, the sensor installed in the manufacturing machine, or the sensor installed in the vicinity of the manufacturing machine in the past and a confirmation result of the operation state of the manufacturing machine at that time, and
update the correlation model such that the error between the correlation model and the correlation feature is reduced, and repeat updating of the correlation model, thereby bringing the correlation model closer to an optimal model according to the environment.
2. The abnormality detector according to claim 1 , wherein
the abnormality detector is configured to detect a sign of an abnormality for each component of the manufacturing machine,
the processor is configured to:
learn an operation state of a component of the manufacturing machine with respect to a physical quantity acquired from the component to be detected for the sign of the abnormality, a sensor installed in the component or a sensor installed in a vicinity of the component,
learn an operation state of a further component of the manufacturing machine with respect to a physical quantity acquired from the further component to be detected for the sign of the abnormality, the sensor installed in the component or the sensor installed in the vicinity of the component, and
estimate the operation states of the component and the further component, using a learning result, based on the physical quantity related to the observed operation of the manufacturing machine, and output the estimation result; and
wherein, even when a sign of an abnormality in the manufacturing machine has been detected, the processor is configured to still suspend output of the signal, according to the output estimation result of the operation state of the manufacturing machine, and an estimation result of the operation state of the manufacturing machine output by another abnormality detector.
3. The abnormality detector according to claim 1 , wherein
from the physical quantity related to the operation of the manufacturing machine, the processor is configured to use a characteristic portion of the physical quantity extracted based on at least one of a fluctuation amount of the physical quantity, an amplitude, a fluctuation time, the number of fluctuations, a frequency, and an amount of deviation from a specified value for outputting a signal indicating an abnormality, for learning.
4. The abnormality detector according to claim 1 , wherein
the processor is configured to classify a waveform detected by the manufacturing machine or detected around the manufacturing machine into any one of
a waveform of a physical quantity indicating the abnormality,
a waveform of a physical quantity indicating stability, and
a waveform of a physical quantity unrelated to the abnormality.
5. The abnormality detector according to claim 2 , wherein, in a case where the operation states of the components estimated by each of the abnormality detectors all affirm the abnormality in the manufacturing machine, the processor is configured to cause to output the signal.