Genetic Programming-based Classification of Ferrograph Wear Particles

Ferrograph analysis is becoming one of the principal methods for condition monitoring and fault diagnosis of the machinery equipment due to its advantages of visualization and efficiency. One of the major challenges of ferrograph analysis is feature construction from the existing features of wear particles to improve classifier efficiency. The current feature construction method is trial and error based on previous experience and mass data, which is time-consuming, laborious and blindness. In this paper, genetic programming-based approach was proposed to construct new features from the five existing morphological features of ferrograph wear particles to improve the ability of classification process. The GP-based feature construction approach is used for fault classification of ferrograph wear particles for the first time and the results show that the method can be used in wear condition monitoring and fault prognosis of machinery equipment.