A ground condition prediction ahead of tunnel face utilizing time series analysis of shield TBM data in soil tunnel

토사터널의 쉴드 TBM 데이터 시계열 분석을 통한 막장 전방 예측 연구

Abstract

This paper presents a method to predict ground types ahead of a tunnel face utilizing operational data of the earth pressure-balanced (EPB) shield tunnel boring machine (TBM) when running through soil ground. The time series analysis model which was applicable to predict the mixed ground composed of soils and rocks was modified to be applicable to soil tunnels. Using the modified model, the feasibility on the choice of the soil conditioning materials dependent upon soil types was studied. To do this, a self-organizing map (SOM) clustering was performed. Firstly, it was confirmed that the ground types should be classified based on the percentage of 35% passing through the #200 sieve. Then, the possibility of predicting the ground types by employing the modified model, in which the TBM operational data were analyzed, was studied. The efficacy of the modified model is demonstrated by its 98% accuracy in predicting ground types ten rings ahead of the tunnel face. Especially, the average prediction accuracy was approximately 93% in areas where ground type variations occur.

토압식(Earth Pressure-Balanced, EPB) 쉴드 TBM 기계데이터 분석을 통해 토사터널의 특징이 반영된 막장 전방 예측 방법을 제안하였다. 기존에 암반과 토사가 혼합된 복합 지반의 예측에 적용하였던 시계열 분석 모델을 토사터널에 적용가능하도록 수정하였다. 또한 수정된 모델을 사용하여, 토사 종류에 따라 쏘일 컨디셔닝 재료를 선택하는 것이 타당한지 연구하였다. 이를 위해 Self-Organizing Map (SOM) 군집화(clustering) 분석을 수행하였다. 그 결과 무엇보다도 지반타입이 #200체 통과량 35% 기준으로 분류되어야 한다는 것을 확인하였다. 또한 TBM 기계데이터 분석을 통해 수정된 모델이 지반 타입을 예측하는데 사용될 수 있음을 확인하였다. 수정된 기준에 따라 지반 타입을 분류하고 시계열 분석을 수행하면, 10막장 전방 지반에 대해서 98%의 높은 예측 정확도를 보였으며, 이를 통해 수정된 방법의 우수성이 입증되었다. 특히 지반 타입 변화 구간에 대한 예측 정확도도 약 93%로, 10막장 전방에서 지반 타입 변화 여부를 미리 확인할 수 있게 되었다.

Fig. 1. Example of deep neural networks

Fig. 2. A time-delay neural network for one-dimensional input/output signals (Hassoun, 1995)

Fig. 3. A result of the SOM clustering

Fig. 4. The profile of machine data groups

Fig. 5. Comparison between the soil conditioning materials and the machine data

Fig. 6. The distribution of the percentage passing through the #200 sieve along the entire route

Fig. 7. Comparison between the soil conditioning materials and the ground types

Fig. 8. Comparison between the soil conditioning materials and the normalized machine data

Fig. 9. Flow chart to run the developed time delay neural network (TDNN) engine (Jung at al., 2018)

Fig. 10. The location of sections at job site

Table 1. Conditions at a job site

Table 2. Properties of soils

Table 3. Classification of ground types

Table 4. Analysis cases

Table 5. TDNN learning method and final model selection criteria (Jung at al., 2018)

Table 6. The results of ANN engine

Table 7. The results of analysis cases

Table 8. Decision criteria for ground type classification in the TDNN model (modified from Jung et al. (2018))

Table 9. Comparison of predicted and actual ground types

Table 10. Comparison of average prediction accuracy

Table 11. Soil conditioning materials according to predicted ground types