Fusion of Neural Networks, Fuzzy Systems and Genetic Algorithms: Industrial Applications Fusion of Neural Networks, Fuzzy Systems and Genetic Algorithms: Industrial Applications
by Lakhmi C. Jain; N.M. Martin
CRC Press, CRC Press LLC
ISBN: 0849398045   Pub Date: 11/01/98
  

Previous Table of Contents Next


The rotating machine diagnosis system performs the feature selection of the fault through theoretical and phenomenological analysis. The mechanical vibration data is transformed by Fourier-Transform, and from its power spectrum, we select the components corresponding to the rotating frequency, to the frequency corresponding to Equations (1) ~ (3), and to some multiple of them (refer to Figure 4).


Figure 4  The feature selection from the power spectrum.

In order to detect the rolling element bearing defect, frequency analysis of the enveloping filter (Figure 5) is used. This method has a low detection accuracy for detecting unbalance and misalignment conditions, but is useful for detecting a bearing defect, which has impulse vibrations represented by Equation (1) ~ Equation (3) (refer to Table 1).

These selected features are diagnosed by the fault diagnosis method using the causal matrix, and the cause of fault is estimated. The causal matrix is shown in Table 2, which represents the correspondence of the cause of fault and the selected feature.

Selected power spectra are used for the estimation of the type of fault using the diagnosis algorithm shown in Figure 6. The diagnosis algorithm performs the multiplication and summation for the selected feature power spectra and the causal matrix parameters. Moreover, it can estimate the type of fault using the total calculated value for the fault. The fault with the highest total calculated value point is evaluated with the highest certainty as the cause of the fault.


Figure 5  Signal processing method of the envelope filter.

Table 1 Corresponding to the type of fault and the feature selection method
The type of fault Feature selection method
FFT Envelope filter + FTT
Unbalance
Misalignment
Bearing inner race defect
Bearing outer race defect
Bearing ball defect
Lack of oil

: Detectable with high accuracy :Detectable with low accuracy : Not detectable

Table 2: An example of the causal matrix
The type of fault 0~fs fs 2fs 3fs fi fo 3fs~ 8fs 8fs~
Unbalance 0 80 20 0 0 0 0 0
Misalignment 0 20 80 0 0 0 0 0
Bearing inner race defect 0 0 0 0 100 0 0 0
Bearing outer race defect 0 0 0 0 0 100 0 0
Lack of oil of bearing 0 0 0 0 0 0 50 50


Figure 6  The fault diagnosis using the causal matrix.

One of the problems in using a causal matrix is dependence on the matrix parameters, because the diagnosis accuracy depends on them. Therefore, in order to improve the diagnosis accuracy, we need some method for optimizing parameters based on experimental fault data.

3. Application of Neural Networks and Fuzzy Logic for Rotating Machine Diagnosis

As mentioned above, in rotating machine diagnosis, the fault diagnosis is performed using selected features from the vibration data. Fault detection is generally performed by the simple judgment method using the absolute threshold value, and the fault identification is performed by a linear classifier using the causal matrix.

In the fault diagnosis, diagnosis accuracy depends upon the suitability of the selected feature and the accuracy of the classifier. The suitability of the selected feature is important to detect the incipient fault, whereas, the performance of the classifier is important to estimate the type of the fault with high certainty.

In this section, we present several examples of using neural network and fuzzy logic applied to the feature selection and the fault identification methods in the rotating machine diagnosis. Neural networks can learn the vibration data from several fault conditions, and we can construct a highly convenient diagnosis system with high accuracy using a neural network. Moreover, we can construct a high reliably diagnosis algorithm using fuzzy logic which can treat vagueness. In the following section, we briefly describe these techniques.


Previous Table of Contents Next

Copyright © CRC Press LLC