A fast fault diagnosis method for rolling bearings, based on extreme learning machine (ELM) and logistic mapping.

Considering both accuracy and the real-time requirement, we propose a novel fast fault diagnosis method for rolling bearings. First, we extract 14 kinds of time-domain features from the original vibration signals and adopt the sequential forward selection (SFS) strategy to select features to ensure a further reduction in dimensionality. Next, we utilize logistic-ELM for fast fault classification, and replace the random input weights in ELM by the logistic mapping sequence.

Extract14Features.m: extract 14 features from the sample data.

FeatureSelection.m: select features which have best performance for fault diagnosis by the sequential forward selection (SFS), and combine into the feature matrix.

FeatureExtractor.m: extract the feature matrix from the sample data.

LogisticMap.m : generate the input weights of ELM by Logistic Mapping.

Logistic_ELM.m : diagnose fault type from the feature matrix.

MainLogisticELM.m : the main funcion, diagnose fault type from the sample data.

We use the rolling bearing vibration signal dataset prepared by the Case Western Reserve University (CWRU) Bearing Data Centre, and you can get if from http://csegroups.case.edu/bearingdatacenter/home.

We also upload the preprocessed data files here as examples, named Data_File.