Automated health state classifier for rotating machines based on expert knowledge
AssigneeSiemens Aktiengesellschaft
InventorsTobias Becker, Jonas Deichmann, Eugen Graz, Henning Ochsenfeld, Thorsten Reimann, Jürgen Zettner
Life of this patent
Estimating the term…
Abstract
The present disclosure relates to a computer-implemented method of, a data processing system for, and a computer program product for indicating machine failures as well as to a corresponding machine and a computer-implemented method of training a neural network for indicating machine failures. At least one input signal based on at least one physical quantity of at least one machine part is transformed into at least one feature. A neural network predicts a class and/or a severity of at least one machine failure based on the at least one feature.
Claims19
A computer-implemented method of indicating machine failures, the method comprising:
- receiving at least one input signal based on at least one physical quantity of at least one machine part of a machine, monitored by at least one sensor;
- transforming the at least one input signal into at least one feature having a different domain and/or a different reference value than the at least one input signal, wherein the at least one input signal is transformed into a spectrum of the at least one input signal and into stochastic features for predominant parts of the spectrum;
- and
- predicting a class and/or a severity of at least one machine failure based on the at least one feature by a neural network trained on specific patterns of the at least one feature or combinations of patterns of more than one feature indicative of specific machine failures.
A data processing system for indicating machine failures, the data processing system comprising:
- a first system having at least one communication unit configured to receive at least one input signal based on at least one physical quantity of at least one machine part of a machine, monitored by at least one sensor;
- a second system having at least one processor configured to transform the at least one input signal into at least one feature having a different domain and/or a different reference value than the at least one input signal, wherein the at least one input signal is transformed into a spectrum of the at least one input signal and into stochastic features for predominant parts of the spectrum;
- and
- a neural network configured to predict a class and/or a severity of at least one machine failure based on the at least one feature, wherein the neural network is trained on specific patterns of the at least one feature or combinations of patterns of more than one feature indicative of specific machine failures.
(canceled)
A machine comprising:
- at least one machine part;
- and
- at least one sensor configured to generate at least one input signal by monitoring at least one physical quantity of the at least one machine part,
- wherein the at least one input signal is provided to a data processing system configured to:
- transform the at least one input signal into at least one feature having a different domain and/or a different reference value than the at least one input signal, wherein the at least one input signal is transformed into a spectrum of the at least one input signal and into stochastic features for predominant parts of the spectrum;
- and predict, by a neural network of the data processing system, a class and/or a severity of at least one machine failure based on the at least one feature, wherein the neural network is trained on specific patterns of the at least one feature or combinations of patterns of more than one feature indicative of specific machine failures.
A computer-implemented method of training a neural network for indicating machine failures, the method comprising:
- generating at least one set of at least one training feature derived from at least one training input signal based on at least one physical quantity of at least one machine part of a machine, wherein specific patterns of the at least one training feature or combinations of patterns of more than one training feature are indicative of specific machine failures, and wherein the at least one training input signal is labelled with known classes and known severities of respective machine failures;
- and
- training the neural network using the at least one set of the at least one training feature.
Dependencies are inferred from the claim text.