Research On Blockchain And Federated Learning Integrated Approach For Failure Detection

In the Industrial Internet of Things,deep learning-based failure detection approaches require enterprises and individuals to share data so as to obtain a large amount of data for model training.With the increasing attention of the society to data privacy,enterprises and individuals are reluctant to share their own data,and the existing failure detection approach begin to face the problem of "data hunger".In the case of less effective data,the generalization ability of the model trained by a single data organization is poor,and it is difficult to be applied to equipment fault detection.As an emerging distributed machine learning,federated learning does not require clients to upload raw data,coordinates clients to jointly train the model,protects the privacy of clients’ data and reduces communication cost,so it is more suitable for the industrial Internet of Things.However,federated learning faces security issue of client data local storage and the difficulty of low user participation to sustain long-term training.Blockchain technology,as a decentralized technology,solves the network trust problem of each node through the consensus mechanism,and ensures the immutability and traceability of data on the chain.Blockchain technology can be used as a special component to integrate in federated learning and provide distributed ledger technology for storing and verifying data for clients in federated learning.This paper studies blockchain and federated learning integrated approach for failure detection,proposes a new failure detection approach,designs an overall framework of the failure detection approach,and provides guidance for the design and implementation of the failure detection approach in the industrial Internet of Things.In the blockchain and federated learning integrated approach,clients train model jointly for failure detection without sharing raw data,which solves the problem of poor generalization ability of the model caused by insufficient effective samples.Through data anchoring algorithm,collected data is used to create the Merkle tree and anchor the root of Merkle tree to blockchain at a predefined time interval.Data anchoring algorithm realizes the verifiable integrity of local data and solves data tampering and other security issues faced by client.In view of the problem that the client is difficult to participate in federated learning for a long time,this paper designs an incentive mechanism based on smart contract.According to the completion status of local training tasks and training contributions,the client is rewarded accordingly to promote the client to participate stably in federated learning.At the same time,the immutability of blockchain data is used to realize the fair distribution of client rewards and prevent malicious theft of tokens.Finally,the proposed approach is implemented in this paper.The accuracy of failure detection approach,and the performance of data anchoring and incentive mechanism are evaluated in bearing failure detection case which is selected as a practical case.Experiment results show that,compared with the model trained by a single client,the recall rate and accuracy of the proposed approach are increased by 1.88% and 1.91% on average.Data anchoring and incentive mechanism have better performance.