The development of an Expert System for TBM application in soft soil

1998. The development of an Expert System for TBM application in soft soil / by A.M.van Noort. ABSTRACT, supervisor: Hack, M.T.J.H. Smits (Fugro Engineers, Leidschendam, The Netherlands), & C.N.Bremmer (NITG, Delft, The Netherlands).  thesis availability: Technical University Delft.

ABSTRACT

The construction of underground infrastructure in the Netherlands was mainly performed with cut-and cover techniques. In a densely populated area like the Netherlands, this technique is complicated to apply. To resolve the problem of building underground with minimum disturbance on the surface, large research projects are initiated to investigate Tunnel Boring Machine (TBM) techniques.

The TBM technology depends on the expertise of many professions, e.g. mining, civil, mechanical engineering as well as geological knowledge, hydrological and economical expertise. To advise the engineer in the pre-design phase of the TBM project a Decision Support System is developed. The Decision Support System exist of an Expert System with rules, definitions and calculations in relation to TBMs and the loose coupling to a 3D-GIS application. The latter is used for interpretation and modeling of the geotechnical and geological information. The Expert System developed for this purpose did not give fully satisfactory results so a new investigation was initiated.

The objective of the research done for this thesis is twofold:

1.     Solve the visibility problem in the alignment choice routine of the existing Expert System for TBMs in soft soil application. Investigate if the utility of a probabilistic approach is applicable.

2.     Reduce the calculation time necessary to calculate the alignment choice routine for a selected TBM. Subsequently design a new expert system, which solves these two problems.

A system has been designed, which is able to calculate the alignment with the highest suitability for a particular TBM in a particular geological environment. The visibility problem of the previous Expert System is overcome by sampling over the entire grid-model. To avoid a loss of reliability by reducing the original grid-model with a large factor, expert knowledge is inserted. A probabilistic approach is investigated and a new deterministic method is researched.

While standard statistics were not found to be useful, two new methods were investigated; the bandwidth and Jack-knife method. By comparing the geology with reference deposition environments, these approaches depend on the determination of the variation of a main TBM parameter, the grainsize. The median grainsize variation could then be labeled to the test-data and the reliability could be calculated. The research showed that the methodology was not suitable and was abandoned.

The developed deterministic approach determines the ‘best’-alignment with a reduced number of grid-cells. Because the reduced number of grid-cells are selected evenly over the model over the entire depth, the visibility problem is solved. Due to a smaller number of sampling points the calculation times are reduced. The remaining problem was the reliability of the results as the density of the input data was extremely decreased. In the end, implementing a new option in Noort’s Expert System, the homogeneity determination, solved this problem. This option calculates whether the first vertical surface is the equal in geometry of applicability as the following. The analysis is performed for the entire model and a table is the output result. The user can select the necessary reduction factor in a specific homogeneous or heterogeneous area. In a homogeneous area only one sampling point is necessary while the rest has the same geometry in applicability. This option enables the Expert System to supply a maximum reliability with a minimum calculation time.

Two different approaches have been investigated and a flow chart is presented to clarify the research (figure 0.1). The main conclusions are that the probabilistic approaches did not give satisfactory results, but the deterministic method supplied remarkable results. The latter approach is successfully implemented in Noort’s Expert System.