A health-adaptive time-scale representation (HTSR) embedded convolutional neural network for gearbox fault diagnostics

•We propose a novel health-adaptive time-scale representation (HTSR) embedded convolutional neural network (HTSR-CNN).•HTSR is constructed to include the enriched fault-related information using multiscale convolutional filters.•The multiscale filters can adaptively capture the fault-related signals through end-to-end learning of the HTSR-CNN.•Experimental results demonstrate the superiority of the proposed HTSR-CNN method for gearbox fault diagnostics.