Incorporating distributed fibre optic sensing and machine learning into dry deep mixing-Paper

Paper

Pdf Summary

The study examines how distributed fibre optic sensing (DFOS) and machine learning (ML) could improve the design and monitoring of dry deep mixing (DDM), a technique used to strengthen soft soils. Conventional verification methods provide limited spatial and temporal data, while current design often depends on costly laboratory and field trials.<br /><br />Laboratory and full-scale field tests evaluated DFOS for measuring strain in stabilised soil columns. Laboratory tests revealed challenges transferring strain from the soil to the stiffer fibre-optic cable, partly because the specimens provided insufficient bonding length. In field trials, DFOS recorded strain changes over time and with depth under an embankment load; measurements showed reasonable agreement with analytical predictions. However, data noise and difficulties ensuring reliable cable installation and bonding remain important limitations.<br /><br />The study also applied random forest and gradient boosting models to predict the undrained shear strength of laboratory-stabilised Norwegian clays. Using soil characteristics, binder contents and curing time as inputs, both models achieved test-set R² values above 0.85 and performed better than traditional predictions based on water-to-binder ratios. Neither algorithm was statistically superior to the other.<br /><br />Overall, DFOS shows promise for providing detailed, repeated measurements of DDM field performance, while ML could improve strength estimates and potentially reduce the need for extensive preliminary testing. Further work is needed to improve sensor installation and strain transfer, identify the most influential ML inputs, tune model parameters and test additional algorithms.

Keywords

distributed fibre optic sensing

machine learning

dry deep mixing

soft soil stabilization

strain monitoring

undrained shear strength

random forest

gradient boosting

field trials

laboratory testing