BTAS: Binary Temporal-Augmented Architecture Search and Application in Rotating Machinery Prognostics and Health Management

Rotating machinery is widely utilized as a core power component in critical equipment, and its health status is crucial for the safe and reliable operation of industrial applications. While deep learning methods have made significant strides, their high computational demands conflict with the resource constraints of industrial edge, where rotating machinery typically operates. To address this issue, we propose a binary trend-aware architecture search (BTAS) strategy for prognostics and health management of rotating machinery. First, a binary network search framework is proposed to compress the prediction model. Second, the search space is refined to minimize information loss due to binarization. Finally, the search target and strategy are optimized to improve prediction performance. The effectiveness of BTAS is demonstrated through two key tasks: remaining useful life prediction and fault diagnosis. BTAS achieves approximately 30-fold memory compression and a 26-fold improvement in computational efficiency while maintaining competitive accuracy.