With the rise of the Internet of Things (IoT), many IoT end devices have entered the market, most of which have the disadvantage of having very constrained computing and storage resources of their own, so the lightweight MQTT protocol is widely used. However, lightweight also means a certain lack of security, so IoT sensors using the MQTT protocol inevitably become the target of many malicious attackers. Digital forensics, as a discipline that applies scientific investigation to digital crime and attacks, can be used to collect evidence of a crime and study how perpetrators attack to in turn help build more dynamic defenses against IoT attacks. However, most current IoT forensic solutions suffer from a lack of granularity and system integrity, so this paper designs a blockchain-based MQTT IoT sensor forensic system that implements a complete path from evidence collection to evidence protection, to evidence analysis and classification, and uses federated blockchains to maintain evidence integrity, and finally the system uses machine learning to qualitative, managing the level of harm to the sensor, forming feedback that can be used to build contingency measures against threats and facilitate the rational allocation of back-end monitoring resources.