To address the challenges of weak early-stage failure features in rotating components under noise interference—where failure features are easily overwhelmed, making extraction difficult—this study employs multiscale feature extraction and proposes a failure diagnosis method based on the Gram Angular Difference Field-Multiscale Convolutional Neural Network (GADF-MCNN) and the Frost-Ice Algorithm-Optimized Kernel Extreme Learning Machine (RIME-KELM). First, GADF is used to convert one-dimensional vibration signals into two-dimensional images to obtain signal information at different scales; second, the images are cropped and compressed before being fed into the MCNN for multiscale feature extraction; finally, the KELM fault diagnosis model is optimized using the RIME algorithm. Using a gear as an example for validation, experimental results show that this method can effectively identify subtle fault features, achieving a fault diagnosis accuracy of 99.2%.To address the challenges of weak early-stage failure features in rotating components under noise interference—where failure features are easily overwhelmed, making extraction difficult—this study employs multiscale feature extraction and proposes a failure diagnosis method based on the Gram Angular Difference Field-Multiscale Convolutional Neural Network (GADF-MCNN) and the Frost-Ice Algorithm-Optimized Kernel Extreme Learning Machine (RIME-KELM). First, GADF is used to convert one-dimensional vibration signals into two-dimensional images to obtain signal information at different scales; second, the images are cropped and compressed before being fed into the MCNN for multiscale feature extraction; finally, the KELM fault diagnosis model is optimized using the RIME algorithm. Using a gear as an example for validation, experimental results show that this method can effectively identify subtle fault features, achieving a fault diagnosis accuracy of 99.2%.To address the challenges of weak early-stage failure features in rotating components under noise interference—where failure features are easily overwhelmed, making extraction difficult—this study employs multiscale feature extraction and proposes a failure diagnosis method based on the Gram Angular Difference Field-Multiscale Convolutional Neural Network (GADF-MCNN) and the Frost-Ice Algorithm-Optimized Kernel Extreme Learning Machine (RIME-KELM). First, GADF is used to convert one-dimensional vibration signals into two-dimensional images to obtain signal information at different scales; second, the images are cropped and compressed before being fed into the MCNN for multiscale feature extraction; finally, the KELM fault diagnosis model is optimized using the RIME algorithm. Using a gear as an example for validation, experimental results show that this method can effectively identify subtle fault features, achieving a fault diagnosis accuracy of 99.2%.

rotating components; multiscale feature extraction; faults diagnosis