Yo-Ming Hsieh

ESFM: An Essential Software Framework for Meshfree Methods

Abstract This paper describes an Essential Software Framework for Meshfree Methods (ESFM). Through thorough analyses of many existing meshfree methods, their common elements and procedures are identified, and a general procedure is formulated into ESFM that can facilitate their implementations and accelerate new developments in meshfree methods. ESFM also modulates performance-critical components such as neighbor-point searching, sparse-matrix storage, and sparse-matrix solver enabling developed meshfree analysis programs to achieve high-performance. ESFM currently consists of 21 groups of classes and 94 subclasses, and more algorithms can be easily incorporated into ESFM. Finally, ESFM provides a common ground to compare various meshfree methods, enabling detailed analyses of performance characteristics.

Cloud-Computing Based Parameter Identification System - with Applications in Geotechnical Engineering

Parameter or system identification is a useful tool in civil engineering. They have wide applications across many disciplines. However, due to the lack of generalpurpose tools for such analyses, researchers need to develop their own codes with duplicated efforts. Yet, it is sometime difficult to adopt other people’s code due to different modeling methods used. Thus, there is a need for general-purpose tools to resolve problems of this kind. This paper describes a general-purpose system for such purpose. The system has three features. First, it uses cloud-computing technology to provide large-scale capabilities. Second, the system uses a JavaScript engine for developing extensions. Third, it employs multi-objective particle swarm optimization for its applicability on wide range of problems. These features allow the system provide general capabilities to solve problems with similar needs. The design is validated and demonstrated by identifying constitutive model parameters in geotechnical engineering.

Observation and Mapping of Complex Landslides Using Field Investigation and Remote Sensed Data

Landslide is always not hazard until mankind development in highly potential area. The study not only tries to map deep seated landslide but also monitoring before the initiation of landslide. Study area in central Taiwan is selected and the geological condition is slate. Major direction of bedding in this area is northeast and the dip ranges from 30° to 75° to southeast. Several deep seated landslides were discovered in the same side of bedding from rainfall events. Several methods such as DEM observation, optical image identification, DInSAR, and monitoring instrumentations are adopted in this study. Landslides in study area are easy to observed from DEM observation and previous traditional elevation survey. However, the activity of landslides should be carefully taken into consideration. Thus SAR data utilization is adopted in this case. DInSAR and SBAS sar analysis are used in this research and ALOS/PALSAR from 2006 to 2010 is adopted. DInSAR analysis shows that landslide is possible mapped but the error is not easy to reduce. The error is possibly form several conditions such as vegetation, clouds, vapor, etc. To conquer the problem, time series analysis, SBAS, is adopted in this research. The result of SBAS in this area shows that large deep seated landslides are easy mapped and the accuracy of vertical displacement is reasonable. Moreover, a landslide area is selected to verify landslide activity in study area. Mems accelerometer is designed to monitor both static and dynamic behavior of this study area. Several events have been observed not only from rainfall but also from earthquake.

How Small Strain Stiffness and Yield Surface Affect Undrained Excavation Predictions

AbstractThis paper discusses how small-strain stiffness and some aspects of the yield surface affect the finite-element predictions of excavation responses, including wall and ground deformations. By using the in-house-developed constitutive model SC1SS, it is possible to isolate and quantify impacts brought about by selected aspects of the soil behavior. This soil model has been validated to reasonably model Taipei silty clay and deep excavations in such soil. This study shows that ignoring small-strain stiffness can overestimate deformations by as much as 80%, leading to conservative and costly design. Inclined yield surface and Lode-angle dependency (on deviatoric planes) have lesser effects on the prediction results. In addition, the mechanisms behind these impacts are investigated through numerical triaxial tests.

Using Motion Sensor for Landslide Monitoring and Hazard Mitigation

Owing to tectonic collision, about 70% of Taiwan’s land are mountains and foothills. With several unfavourable conditions such as typhoons, strong rainfalls, and earthquakes, the landslide hazard is a critical issue in Taiwan. This issue is increasingly important because the use of hillsides for residence are growing in recent years. To address the issue and to protect its citizens, Taipei City Government initiated a pilot project to monitor landslide displacements in 2010. The project involves a selected hillside community with dip slope in Taipei City. The community had a major landslide in 1983.The monthly monitoring with non-automated devices is not helpful for residence to react to landslides. To help improve the situation, we use motion sensors to develop fully automated devices to infer displacements. We also used numerical analyses to help determine the best locations to install these devices. They were mounted to the ground surface to sense the tilt. The tilt converts to displacements by assuming slip surfaces. Small displacements were detected during a typhoon event in July of 2013.With the developed monitoring device, standard operation procedures were developed accordingly. Community residents has been educated how to react to unfavourable conditions. These conditions are defined by displacements exceeding certain thresholds.

Localization inside Tunnels Using Machine Vision

Localization (position tracking) inside tunnels is difficult. GPS signal is not available inside tunnels, and installation of wireless access points is expensive. In this work, we propose the use machine vision for localization in tunnels. The advantage for such approach is that it requires no installation of any infrastructure, needed with radio triangulation approach. Prior researches have shown that machine vision can successfully identify positions in both outdoor and indoor environments. However, such application has never been tried inside tunnels. In this paper, we present our work on developing appropriate algorithms for localization in tunnels. One potential application for localization in tunnels is to help tunnel maintainers acquire accurate location information in tunnels when they do tunnel

Computer Aided Iterative Design – A Future Trend in Computer Aided Engineering Software

Typical engineering design processes involve an initial design, followed by iterations of analyses, interpreting and evaluating analyses results, and proposing new design or modifying existing design. The final design often results from intuitions obtained during this iterative process. Most computer-aided engineering (CAE) software focuses on the computer-aided analysis by advancing capabilities in pre-processing, computation, and post-processing. However, little attention has been paid to the support of the aforementioned iterative design process, which is the common practice in engineering design. Authors believe the next-generation CAE software should evolve into CAID (ComputerAided Iterative Design) software to help engineers go through the iterative design process and develop better engineering designs. In this paper, key software requirements for the iterative design process are discussed in this work. Furthermore, prototype CAID software developed using C, Qt, and VTK is demonstrated. The analyzing capability of the developed CAID software is based on an essential software framework for meshfree methods (ESFM). The proposed CAID concept and prototype software shall provide guidelines for future CAE software development.

An Efficient Scheme for Parallel Parametric-Study in Finite Element Analyses

Parametric studies in parallel-computing is often considered as an “embarrassingly parallel” computation task that can be trivially parallelized and yield good parallel efficiency. However, this paper shows the parallel efficiency of such analyses can be further improved by the proposed scheme in typical cluster setups using commodity PC hardware. Numerical experiments show that the parallel efficiency of such computation task in finite element analyses can be improved by 81.6% and the peak disk usage can be reduced by 83.1%. The proposed scheme for enhancing parallel performance can be readily extended to design optimization problems, which has similar work patterns in parallel processing. Therefore, more efficient engineering analyses and designs can be achieved by employing schemes described in this work.

An Object-oriented Framework for Spatial Interpolation

Interpolation is an important operator in numerical methods for solving partial differential equations and in geospatial applications. There are many interpolation methods proposed in the past. In this work, a unified software framework is proposed through the use of design-patterns in object-oriented programming. By using this framework, little effort is necessary to implement different interpolations algorithms when commonality with implemented algorithms can be found. Furthermore, through this framework, it becomes easy to compare the performance of different algorithms because of the unified application interface..

Efficient Computation for a Complex Tunnel Excavation Problem

Computational efficiency is of critical importance in the application of large-scale finite element simulations for solving complex geotechnical problems. This paper summarizes the formulation and implementation of a FETI domain decomposition technique for parallel finite element computations, highlighting the parallel efficiency and scalability of the software using a cluster of 16 interconnected commodity computers. The program is then used to simulate ground movements caused by the construction of a shallow cavern for a new subway station using a stacked-drift tunneling method which involved hand-excavation of a sequence of 15, 3m-square-section drifts through weathered alluvial soils. Figure 1 shows the cross-section of the cavern for the recently completed Rio Piedras station, part of the new Tren Urbano system in San Juan, Puerto Rico. The horseshoe-shaped cavern (17m wide, 16m high, and 150m long) was constructed in weathered alluvial soils. The crown of the cavern was located less than 5.5m below existing build- ings in a busy historic shopping district. Structural support for the cavern was provided by a series of 15 stacked drifts. These rectangular-section galleries (each typically 3m square) were excavated mainly by hand between two 30m deep access shafts and were then in-filled with concrete, while a primary compensation grouting system was designed to mitigate effects of excavation-induced ground movements on the overlying structures. Unexpectedly large settlements occurred during the early stages of drift construction and overwhelmed the original grouting sys- tem. As a result, an extensive consolidation grouting system was installed (to enable grouting closer to the tunnel heading). Despite these remedial measures, building settlements above the cavern at the end of construction ranged from 70mm to 130mm. In principle, non-linear finite element methods can be used to evaluate/predict the ground deformations caused by the stacked-drift construction. However, the complexity of the project presents a major challenge both in details of the modeling and in the scale of the computation. A comprehensive numerical model of the cavern would require detailed modeling of the unusual engineering properties and spatial variability of the weathered alluvium (Zhang et al. 2003), simulation of grout injections (and their effects on the surrounding soil), as well as the primary details of the drift excavation, lining and concrete infilling for each of the 15 drifts. The actual construction proce- dure involved concurrent excavation of between two and four drifts, Figure 1b (sometimes advancing in opposite directions), while concrete infilling took place from 3 to 60 days after excavation. Preliminary calculations found that very large 3-D finite element models (O(10 5 ) degrees of freedom) would be necessary to represent even these basic details of the stacked-drift construction. When non-linear soil behavior (with marginal face stability) was in- cluded in the finite element analyses, it soon became apparent that solutions using conventional direct equilibrium equation solvers or iterative (Krylov sub-space etc.) methods with single-processor machines would exceed practical limits on computational time and hence, a more efficient approach was required.