US researchers say a self-supervised machine-learning tool can identify long-term physical defects in solar assets weeks or years before conventional inspections, potentially reducing operations and ...