Abstract:

The mudcake incidents would significantly reduce the effectiveness of shield building and jeopardize worker safety when an earth pressure balance (EPB) shield is being built. To overcome the challenge of recognizing mud cake events in shields, a method is proposed to diagnose mud cake events from two perspectives of data modeling and knowledge inference. Then, the mudcake diagnosis submodel based on unthresholded recurrence plot transformation and extreme learning machine autoencoder technology in data modeling is established. The mudcake event rules are also retrieved from a knowledge reasoning standpoint to create an event diagnostic knowledge base and choose a mudcake event diagnosis submodel based on prior construction experience. Finally, based on the two submodel output results and the comprehensive probability method, a mudcake event early warning model powered by data and knowledge is built. The use of the early warning model demonstrates that the mudcake event early warning model driven by both data and knowledge model can accurately predict the mudcake event with better performance and lower false positive rate than single submodels. Furthermore, the warning time is earlier than the manual discovery time, providing technical help for project safety.

Key words: mud cake, earth pressure balance shield, datadriven; knowledgedriven, anomaly detection, early warning model