Fault prediction model based on evidential reasoning approach

Abstract

In order to deal with fault prediction problems that involve both quantitative and qualitative information for nonlinear complex system, a new fault prediction model is established based on the evidential reasoning (ER) approach, and an optimal learning algorithm for training ER-based prediction model is presented based on the mean square error (MSE) criterion. This prediction model inherits the advantages of ER approach, which can deal with precise data, incomplete data and fuzzy data with nonlinear characteristic. In this model, the input signals transformed using rule based information transformation technique, are aggregated by analytical ER approach, and then the outputs of prediction model are constructed according to the types of system outputs. In addition, two fault decision criteria are defined to conduct fault identification. To overcome the difficulty in determining model parameters accurately and subjectively, a nonlinear optimization model is constructed and the optimal parameters are obtained. Two experimental studies are conducted to evaluate the performance of the proposed model. The results show that the established prediction model and the presented parameters optimization methods can deal with fault prediction problem effectively.

A Diagnosis Method for Rotation Machinery Faults Based on Dimensionless Indexes Combined withK-Nearest Neighbor Algorithm

“…Based on the collected failure data, dimensionless indexes can be calculated and fault zone (the maximum and minimum range in 10 indices) can be set up. Use (2), (3), (4), (5), and (6) to calculate the waveform indices, peak indicators, pulse index, margin index, and the kurtosis index range faults.…”

“…Solving the above problems requires the use of an effective method which can process uncertain information rationally, systematically, and flexibly [5]. Evidence theory can effectively express and deal with uncertain and imprecise information and other problems [6].…”

Fusion of the Dimensionless Parameters and Filtering Methods in Rotating Machinery Fault Diagnosis

“…In rotating machinery fault diagnosis usually use the time domain or frequency domain analysis of vibration monitoring data for fault diagnosis [2][3][4][5]. Rotating machinery in the event of a failure, however, vibration monitoring signals tend to have a large number of non-linear, random, non-ergodic information, and bring great difficulty in fault signal analysis [6].…”

“…By contrast, the dimensionless index is not sensitive to the disturbance of vibration monitoring signal, performance is stable. In particular, these dimensionless index are not sensitive to the change of amplitude and frequency of the signal, namely, it has little relationship with working conditions of the machine [1][2][3][5][6][7]. Therefore, the dimensionless index has been widely used in fault diagnosis of rotating machinery.…”

“…In dimensionless index, pulse index and kurtosis index is more sensitive to impact type fault, especially in the early failure, the large amplitude of the pulse is less, other parameter values increase is not much, but kurtosis index and pulse index rise faster, so that failure of the range is larger. It is difficult to determine the scope of the composite fault interval [1][2][3][5][6][7].…”