Fault diagnosis of rolling bearing based on permutation entropy and Extreme Learning Machine

In consideration that bearing faults usually appear in the form of periodical impacts, fault information is easily interfered by noises and it is difficult to extract fault features, a method is proposed for diagnosing faults of roller bearings based on PE (Permutation Entropy) and ELM (Extreme Learning Machine). First of all, signals of original acceleration and vibration are decomposed at several levels for bearings through MRSVD (Multi-resolution Singular Value Decomposition) to extract detailed components including fault features. Subsequently, PE values are extracted from the detailed components as values of fault characteristics, in order to construct feature vectors and input them to the ELM for training and recognition. This method is used for diagnosing local faults of roller bearings in their balls, inners and outers in practices, by analyzing the experimental data, the precision of fault recognition is up to 97.5% and thus proves that this method is highly effective for diagnosing faults of the bearings.