Industrial　Internet　of　Things　(IIoT)　is　widely　applied　in　modern　industries.　However,　IIoT　faces　challenges　in　system　modeling　and　cyber　security.　Distributed　data　isolation　influences　the　time　series　data　modeling,　while　frequent　communications　over　IIoT　leave　vulnerabilities　to　cyber　attacks.　To　ensure　proper　IIoT　operations,　we　propose　a　federated　temporal　learning　(FTL)　mechanism　to　extract　temporal　features　for　cyber　attack　detection.　We　build　local　FTL　models　to　accurately　extract　local　time　series　features　on　edge　devices.　We　collaboratively　construct　the　global　FTL　model　to　solve　the　distributed　heterogeneity　at　edges　to　further　improve　cyber　attack　detection.　Moreover,　we　build　a　testing　platform　and　emulate　realistic　cyber　attacks　towards　IIoT.　The　proposed　FTL　detects　seven　types　of　cyber　attacks　more　effectively　compared　with　traditional　models,　in　terms　of　comprehensive　evaluation　metrics.　Thus,　FTL　is　a　promising　IIoT　security　solution　for　distributed　industrial　systems.　©　2023　IEEE.