Transfer between multiple machine plants: A modified fast self-organizing feature map and two-order selective ensemble based fault diagnosis strategy

The signal differences cause machine fault diagnosis (MFD) models developed in one plant to not be readily applicable to others. This paper presents a modified fast self-organizing feature map (FSOM) and two-order selective ensemble (SE) strategy to realize transfer learning (TL) between multiple plants, including three major processes: i) modified FSOM to map the original real-imaginary polar diagrams to a new feature space where the differences in the same fault category are reduced, ii) cross Minkowski distance matrix to calculate the similarity between channels, and to select the helpful channels in the source plant by an evaluation process, iii) two-order SE to fuse high-powered channels in the target plant to promote diagnosis. Experiments in two gearbox systems demonstrate the effectiveness of transferring from a simple/local to a complex/global device, thus being a useful tool to solve the practical problem that model in the laboratory and apply in the industrial field.