In　this　study,　a　non-destructive　testing　method　for　crack　defects　of　self-lubricating　sliding　bearing　coating　based　on　terahertz　time-domain　spectroscopy　was　proposed.　The　self-lubricating　coating　materials　were　revealed　to　have　good　penetration　and　characteristic　response　in　the　terahertz　band　through　experiments.　To　solve　the　problem　of　difficulty　in　signal　feature　extraction　caused　by　overlap,　the　broad　learning　system　was　used　to　classify　and　predict　time-domain　signals　of　crack　defects.　The　identification　accuracy　for　crack　defects　is　96.08%,　and　the　mean　relative　errors　of　prediction　for　interface　and　internal　cracks　(5　to　95　mu　m　in　size)　are　4.16%　and　3.40%,　respectively.　The　method　proved　the　applicability　for　qualitative　and　quantitative　evaluation　in　crack　defects,　which　is　considered　a　new　idea　for　the　non-destructive　testing　of　self-lubricating　coating.