Feature generation in fault diagnosis based on immune programming

In the symptom feature discovery, genetic programming has the shortage of premature convergence. So a new feature generation method based on immune programming is put forward. The new features are constructed by polynomial expressions of the original features. And then, with the immune operators such as antibody representation and mutation of tree-like structure, affinity function defined by classification performance of every individual, the clonal selection optimal algorithm is adopted to search the best feature that has excellent classification performance. The experiments of sound signal for gasoline engine show that, due to the diversity of antibodies is maintained by clonal selection principle, the best compound feature founded by immune programming has better classification ability than feature optimized by genetic programming.