Data processing: artificial intelligence – Neural network – Learning task
Reexamination Certificate
2000-03-02
2003-06-03
Follansbee, John A. (Department: 2121)
Data processing: artificial intelligence
Neural network
Learning task
C706S002000, C706S020000, C706S905000, C706S913000
Reexamination Certificate
active
06574613
ABSTRACT:
FIELD OF THE INVENTION
The invention is directed to a system and a method for diagnosis of engine conditions.
BACKGROUND OF THE INVENTION
Traditionally, the diagnosis of engine conditions utilizes vibration signals on the basis of amplitude limits. The vibration amplitude limits have been derived from general experience and/or on the basis of features from vibration signatures, from the experience deriving from events during the development phase or, respectively, the experience from the certification or testing process.
Costly and time-consuming modifications in mass production of engines usually ensue.
The vibration diagnosis has been implemented by variously qualified specialist teams without a targeted exchange of experience between operators and manufacturers of engines and without systematic acquisition and interpretation of errors, side-effects or, respectively, symptoms and their causes.
In the previously standard vibration diagnosis of engine conditions, there is thus, among other things, the problem that few measuring positions are contrasted to only a limited amount of information for interpretation. There are in fact error catalogs from the development phase; these, however, are usually full of gaps. The influence of a great number of parameters such as, for example, construction standards, tolerances, size and position of unbalanced masses, temperature effects, performance and flight parameters, etc., as well as non-linearities and measuring imprecisions, remain largely unconsidered.
Given this type of vibration diagnosis, dangerous vibration conditions can continue to exist unrecognized during operation. More serious secondary damage due to late recognition can occur and the outlay for maintenance increases since it is usually necessary to dismantle an engine.
SUMMARY OF THE INVENTION
The present invention pertains to systems and a methods for diagnosis of engine conditions. The systems and methods are directed to extraction of features or parameters from different information sources and to processing of the features. These features, together with a series connection of two neural networks, provide a dependable diagnosis of engine conditions, particularly error recognition.
In an embodiment of the present invention, a system for diagnosis of engine conditions has:
a means for supplying statistical/probabilistic information about the error quota of individual engine components resulting from an evaluation of a corresponding data bank and/or
a plurality of measurement sensors for acquiring physical information such as, for example, pressures and temperatures in various engine levels and, moreover, parameters from a particle analysis in used oil and in engine exhaust gases as well as parameters from an analysis of the gas path;
a plurality of measurement sensors for acquiring vibration information in the time domain from an engine;
a vibration analysis means for generating vibration information in the frequency domain from the vibration information in the time domain;
a module for feature extraction for processing the physical information and/or the statistical/probabilistic information and the vibration information in the time and frequency domain and for the extraction of a number of features that comprehensively describe the engine condition;
a first neural network to which the features are applied for classification of the features, for identification of relationships and dependencies between features and for corresponding implementation of an information compression and for output of parameters, whereby the first neural network comprises an input layer, one or more intermediate, layers and an output layer of neurons, whereby the input layer comprises more neurons than the intermediate layer(s) and this in turn comprises more neurons than the output layer, and the neurons of a layer are connected via a plurality of connecting elements having variable weighting coefficients;
a first training means for supplying training input signals to the first neural network and for comparison of the output signal output by the first neural network in response thereto to a training input signal and for the modification of variable weighting coefficients of the first neural network by means of application of a predetermined training algorithm corresponding to the differences between the training input signal and the output signal or for realizing a non-monitored training of the first neural network with the assistance of the training input signals by themselves;
a second neural network to which the parameters output by the first neural network are applied for classification of the parameters, for recognition of relationships between the parameters and specific error constellations, for corresponding implementation of an information linkage and for output of a diagnosis signal, whereby the second neural network comprises an input layer, one or more intermediate layers and an output layer of neurons, whereby the input and the output layer comprise fewer neurons than the intermediate layer(s), and the neurons of a layer are connected to the neurons of the layer following thereupon via a plurality of connecting elements having variable weighting coefficients; and
a second training means for supplying training input signals to the second neural network and for comparing the output signal obtained from the second neural network in response thereto to a training input signal and for modifying variable weighting coefficients of the second neural network by means of applying a predetermined training algorithm corresponding to the differences between the training input signal and the output signal.
In an embodiment of the present invention, the module for feature extraction employs physical parameters such as oil consumption given specific engine runs, power reference numbers such as pressure and temperature in specific engine levels, parameters from a particle analysis in used oil and in engine exhaust gases as well as parameters from an analysis of the gas path.
In an embodiment of the present invention, the module for feature extraction employs methods that are standard for speech recognition, and extracts effective values, properties of the envelopes, modulations, absolute values, performance analyses, statistical parameters, distribution functions, wavelet analysis, etc., of the vibration information in the time domain as features.
In an embodiment of the present invention, the vibration analysis means handles the vibration signals in the time domain and determines corresponding vibration information in the frequency domain therefrom.
In an embodiment of the present invention, the module for feature extraction employs an information presentation in the form of what is referred to as a waterfall diagram, handles this information presentation with image processing methods and determines corresponding features therefrom from the vibration information in the frequency domain.
In an embodiment of the present invention, the module for feature extraction also implements geometrical considerations of the overall image or specific image regions; and/or the module for feature extraction also considers what are referred to as “skylines” of the waterfall diagram from the perspective of the frequency or, respectively, of the time/speed access.
In an embodiment of the present invention, the module for feature extraction also numerically acquires the vibration information of the waterfall diagrams; and utilizes methods from matrix and vector calculation or methods for system identification in the frequency domain for acquiring features from the vibration information in the frequency domain and/or utilizes transfer functions as well as a distribution analysis of the numerical data.
In an embodiment of the present invention, the neural networks in combination with fuzzy logic or pure fuzzy logic circuits are provided instead of the first and second neural networks.
In an embodiment of the present invention, a method for diagnosis of engine conditions has the steps:
supplying statistical/probabilis
Booker Kelvin
Follansbee John A.
LandOfFree
System and method for diagnosing jet engine conditions does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with System and method for diagnosing jet engine conditions, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and System and method for diagnosing jet engine conditions will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3125922