In　Industry　4.0,　with　the　increasing　scale　of　data　generated　in　IIoT,　it　is　necessary　for　federated　learning　(FL)　algorithms　to　process　and　analyze　these　data　in　real　time,　thereby　quickly　generating　high-quality　models　for　edge　computing/intelligence.　But　there　are　still　challenges　on　current　FL　frameworks　in　IIoT,　such　as　difficult　client　management,　prolonged　communication　delays,　and　compromised　learning　effectiveness　induced　by　attacks.　To　address　these　challenges,　we　propose　a　new　FL　framework　that　integrates　a　digital　identity　module　for　user　perception　and　authentication,　a　decentralized　blockchain　module　for　trustworthy　FL,　and　an　adaptive　model　sparsification　algorithm　for　communication-assisted　sensing　FL.　Our　FL　framework　aims　to　conduct　some　sense　tasks　on　image　classification　and　sentiment　analysis.　The　effectiveness　of　our　proposed　framework　for　IIoT　is　demonstrated　through　technical　explanations　and　experimental　results.　©　2023　ACM.