Abstract
Available machine fault diagnostic methods show
unsatisfactory performances on both on-line and
intelligent analyses because their operations involve
intensive calculations and are labour intensive. Aiming
at improving this situation, this paper describes the
development of an intelligent approach by using the
Genetic Programming (abbreviated as GP) method.
Attributed to the simple calculation of the
mathematical model being constructed, different kinds
of machine faults may be diagnosed correctly and
quickly. Moreover, human input is significantly reduced
in the process of fault diagnosis. The effectiveness of
the proposed strategy is validated by an illustrative
example, in which three kinds of valve states inherent
in a six-cylinders/four-stroke cycle diesel engine,
i.e. normal condition, valve-tappet clearance and gas
leakage faults, are identified. In the example, 22
mathematical functions have been specially designed and
8 easily obtained signal features are used to construct
the diagnostic model. Different from existing GPs, the
diagnostic tree used in the algorithm is constructed in
an intelligent way by applying a power-weight
coefficient to each feature. The power-weight
coefficients vary adaptively between 0 and 1 during the
evolutionary process. Moreover, different evolutionary
strategies are employed, respectively for selecting the
diagnostic features and functions, so that the
mathematical functions are sufficiently and in the
meantime, the repeated use of signal features may be
fully avoided. The experimental results are illustrated
diagrammatically in the following sections.
Users
Please
log in to take part in the discussion (add own reviews or comments).