No description available.

The objective of this thesis is to develop and validate a computational framework based on mathematical models for the motion prediction and dynamic stability quantification of human walking, which can differentiate the dynamic stability of human walking with different mechanical properties of the leg. Firstly, a large measurement database of human walking motion was created. It contains walking measurement data of 8 subjects on 3 self-selected walking speeds, which 10 trials were recorded at each walking speed. The motion of whole-body centre of mass and the leg were calculated from the kinetic-kinematic measurement data. The fundamentals of leg property have been presented, and the parameters of leg property were extracted from the measurement data of human walking where the effects of walking speed and condition of foot-ground contact were investigated. Three different leg property definitions comprising linear axial elastic leg property, nonlinear axial elastic leg property and linear axial-tangential elastic leg property were used to extracted leg property parameters. The concept of posture-dependent leg property has been proposed, and the leg property parameters were extracted from the measurement data of human walking motion where the effects of walking speed and condition of foot-ground contact were also investigated. The compliant leg model with axial elastic property (CAE) was used for the dynamic stability analysis of human walking with linear and nonlinear axial elastic leg property. The compliant leg model with axial and tangential elastic property (CATE) was used for that with linear axial-tangential elastic leg property. The posture - dependent elastic leg model (PDE) was used for that with posture-dependent leg property. It was found that, with linear axial elastic leg property, the global stability of human walking improves with the bigger touchdown contact angle. The average leg property obtained from the measurement data of all participants allows the maximum global stability of human walking. With nonlinear axial elastic leg property, the global stability decreases with the stronger nonlinearity of leg stiffness. The incorporation of the tangential elasticity improves the global stability and shifts the stable walking velocity close to that of human walking at self-selected low speed (1.1-1.25 m/s).By the PDE model, the human walking motions were better predicted than by the CATE model. The effective range of walking prediction was enlarged to 1.12 – 1.8 m/s. However, represented by PDE model, only 1-2 walking steps can be achieved. In addition, the profiles of mechanical energies represented by the PDE model are different from that of the orbital stable walking represented by CATE model. Finally, the minimal requirements of the human walking measurements and the flexibility of simple walking models with deliberate leg property definitions allow the computational framework to be applicable in the dynamic stability analysis of the walking motion with a wide variety of mechanical property of the leg.

The purpose of this thesis is to conduct a detailed examination of the forward-directivity characteristics of near-fault ground motions produced in the 2010-11 Canterbury earthquakes, including evaluating the efficacy of several existing empirical models which form the basis of frameworks for considering directivity in seismic hazard assessment. A wavelet-based pulse classification algorithm developed by Baker (2007) is firstly used to identify and characterise ground motions which demonstrate evidence of forward-directivity effects from significant events in the Canterbury earthquake sequence. The algorithm fails to classify a large number of ground motions which clearly exhibit an early-arriving directivity pulse due to: (i) incorrect pulse extraction resulting from the presence of pulse-like features caused by other physical phenomena; and (ii) inadequacy of the pulse indicator score used to carry out binary pulse-like/non-pulse-like classification. An alternative ‘manual’ approach is proposed to ensure 'correct' pulse extraction and the classification process is also guided by examination of the horizontal velocity trajectory plots and source-to-site geometry. Based on the above analysis, 59 pulse-like ground motions are identified from the Canterbury earthquakes , which in the author's opinion, are caused by forward-directivity effects. The pulses are also characterised in terms of their period and amplitude. A revised version of the B07 algorithm developed by Shahi (2013) is also subsequently utilised but without observing any notable improvement in the pulse classification results. A series of three chapters are dedicated to assess the predictive capabilities of empirical models to predict the: (i) probability of pulse occurrence; (ii) response spectrum amplification caused by the directivity pulse; (iii) period and amplitude (peak ground velocity, PGV) of the directivity pulse using observations from four significant events in the Canterbury earthquakes. Based on the results of logistic regression analysis, it is found that the pulse probability model of Shahi (2013) provides the most improved predictions in comparison to its predecessors. Pulse probability contour maps are developed to scrutinise observations of pulses/non-pulses with predicted probabilities. A direct comparison of the observed and predicted directivity amplification of acceleration response spectra reveals the inadequacy of broadband directivity models, which form the basis of the near-fault factor in the New Zealand loadings standard, NZS1170.5:2004. In contrast, a recently developed narrowband model by Shahi & Baker (2011) provides significantly improved predictions by amplifying the response spectra within a small range of periods. The significant positive bias demonstrated by the residuals associated with all models at longer vibration periods (in the Mw7.1 Darfield and Mw6.2 Christchurch earthquakes) is likely due to the influence of basin-induced surface waves and non-linear soil response. Empirical models for the pulse period notably under-predict observations from the Darfield and Christchurch earthquakes, inferred as being a result of both the effect of nonlinear site response and influence of the Canterbury basin. In contrast, observed pulse periods from the smaller magnitude June (Mw6.0) and December (Mw5.9) 2011 earthquakes are in good agreement with predictions. Models for the pulse amplitude generally provide accurate estimates of the observations at source-to-site distances between 1 km and 10 km. At longer distances, observed PGVs are significantly under-predicted due to their slower apparent attenuation. Mixed-effects regression is employed to develop revised models for both parameters using the latest NGA-West2 pulse-like ground motion database. A pulse period relationship which accounts for the effect of faulting mechanism using rake angle as a continuous predictor variable is developed. The use of a larger database in model development, however does not result in improved predictions of pulse period for the Darfield and Christchurch earthquakes. In contrast, the revised model for PGV provides a more appropriate attenuation of the pulse amplitude with distance, and does not exhibit the bias associated with previous models. Finally, the effects of near-fault directivity are explicitly included in NZ-specific probabilistic seismic hazard analysis (PSHA) using the narrowband directivity model of Shahi & Baker (2011). Seismic hazard analyses are conducted with and without considering directivity for typical sites in Christchurch and Otira. The inadequacy of the near-fault factor in the NZS1170.5: 2004 is apparent based on a comparison with the directivity amplification obtained from PSHA.

The characteristics of earthquake shaking are affected by the local site conditions. The effects of the local soil conditions are often quantified via an amplification factor (AF), which is defined as the ratio of the ground motion at the soil surface to the ground motion at a rock site at the same location. Amplification factors can be defined for any ground motion parameter, but most commonly are assessed for acceleration response spectral values at different oscillator periods. Site amplification can be evaluated for a site by conducting seismic site response analysis, which models the wave propagation from the base rock through the site-specific soil layers to the ground surface. An alternative to site-specific seismic response analysis is site amplification models. Site amplification models are empirical equations that predict the site amplification based on general characteristics of the site. Most of the site amplification models that already used in ground motion prediction equations characterize a site with two parameters: the average shear wave velocity in the top 30 m (VS30) and the depth to bedrock. However, additional site parameters influence site amplification and should be included in site amplification models. To identify the site parameters that help explain the variation in site amplification, ninety nine manually generated velocity profiles are analyzed using seismic site response analysis. The generated profiles have the same VS30 and depth to bedrock but a different velocity structure in the top 30 m. Different site parameters are investigated to explain the variability in the computed amplification. The parameter Vratio, which is the ratio of the average shear wave velocity between 20 m and 30 m to the average shear wave velocity in the top 10 m, is identified as the site parameter that most affects the computed amplification for sites with the same VS30 and depth to bedrock. To generalize the findings from the analyses in which only the top 30 m of the velocity profile are varied, a suite of fully randomized velocity profiles are generated and site response analysis is used to compute the amplification for each site for a range of input motion intensities. The results of the site response analyses conducted on these four hundred fully randomized velocity profiles confirm the influence of Vratio on site amplification. The computed amplification factors are used to develop an empirical site amplification model that incorporates the effect of Vratio, as well as VS30 and the depth to bedrock. The empirical site amplification model includes the effects of soil nonlinearity, such that the predicted amplification is a function of the intensity of shaking. The developed model can be incorporated into the development of future ground motion prediction equations. / text

This study evaluates the seismically induced displacements of a slope using the Finite Element Method (FEM) in comparison to the results of twelve empirical predictive approaches. First, the existing methods to analyze the stability of slopes subjected to seismic loads are presented and their capabilities to predict the onset of failure and post-failure behavior are discussed. These methods include the pseudostatic method, the Newmark method, and stress-deformation numerical methods. Whereas the pseudostatic method defines a seismic coefficient for the analysis and provides a safety factor, the Newmark method incorporates a yield coefficient and the actual acceleration time history to estimate permanent displacements. Numerical methods incorporate advanced constitutive models to simulate the coupled stress-strain soil behavior, making the process computationally more costly. In this study, a model slope previously studied at laboratory scale is selected and scaled up to prototype dimensions. Then, the slope is subjected to 88 different input motions, and the seismic displacements obtained from the numerical and empirical approaches are compared statistically. From correlation analyses between seven ground motion parameters and the numerical results, new empirical predictive equations are developed for slope displacements. The results show that overall the FEM displacements are generally in agreement with the numerically developed methods by Fotopoulou and Pitilakis (2015) labelled "Method 2" and "Method 3", and the Newmark-type Makdisi and Seed (1978) and Bray and Travasarou (2007) methods for rigid slopes. Finally, functional forms for seismic slope displacement are proposed as a function of peak ground acceleration (PGA), Arias intensity (Ia), and yield acceleration ratio (Ay/PGA). These functions are expected to be valid for granular slopes such as earth dams, embankments, or landfills built on a rigid base and with low fundamental periods (Ts<0.2). / Master of Science / A landslide is a displacement on a sloped ground that can be triggered by earthquake shaking. Several authors have investigated the failure mechanisms that lead to landslide initiation and subsequent mass displacement and proposed methodologies to assess the stability of slopes subjected to seismic loads. The development of these methodologies has to rely on field data that in most of the cases are difficult to obtain because identifying the location of future earthquakes involves too many uncertainties to justify investments in field instrumentation (Kutter, 1995). Nevertheless, the use of scale models and numerical techniques have helped in the investigation of these geotechnical hazards and has led to development of equations that predict seismic displacements as function of different ground motion parameters. In this study, the capabilities and limitations of the most recognized approaches to assess seismic slope stability are reviewed and explained. In addition, a previous shaking-table model is used for reference and scaled up to realistic proportions to calculate its seismic displacement using different methods, including a Finite Element model in the commercial software Plaxis2D. These displacements are compared statistically and used to develop new predictive equations. This study is relevant to understand the capabilities of newer numerical approaches in comparison to classical empirical methods.

The KiK-net ground motion database is used to develop ground motion prediction equations for Arias Intensity (I_a), 5-95% Significant Duration (Ds_5-95), and 5-75% Significant Duration (Ds_5-75). Relationships are developed both for shallow crustal earthquakes and subduction zone earthquakes (hypocentral depth less than 45 km). The models developed consider site amplification using V_S30 and the depth to a layer with V_S=800 m/s (h₈₀₀). We observe that the site effect for I_α is magnitude dependent. For Ds_5-95 and Ds_5-75, we also observe strong magnitude dependency in distance attenuation. We compare the results with previous GMPEs for Japanese earthquakes and observe that the relationships are similar. The results of this study also allow a comparison between earthquakes in shallow-crustal regions, and subduction regions. This comparison shows that Arias Intensity has similar magnitude and distance scaling between both regions and generally Arias Intensity of shallow crustal motions are higher than subduction motions. On the other hand, the duration of shallow crustal motions are longer than subduction earthquakes except for records with large distance and small magnitude causative earthquakes. Because small shallow crustal events saturate with distance, ground motions with large distances and small magnitudes have shorter duration for shallow crustal events than subduction earthquakes.

The vast majority of products and processes in industry and academia require human interaction. Thus, digital human models (DHMs) are becoming critical for improved designs, injury prevention, and a better understanding of human behavior. Although many capabilities in the DHM field continue to mature, there are still many opportunities for improvement, especially with respect to posture- and motion-prediction. Thus, this thesis investigates the use of artificial neural network (ANN) for improving predictive capabilities and for better understanding how and why human behave the way they do. With respect to motion prediction, one of the most challenging opportunities for improvement concerns computation speed. Especially, when considering dynamic motion prediction, the underlying optimization problems can be large and computationally complex. Even though the current optimization-based tools for predicting human posture are relatively fast and accurate and thus do not require as much improvement, posture prediction in general is a more tractable problem than motion prediction and can provide a test bead that can shed light on potential issues with motion prediction. Thus, we investigate the use of ANN with posture prediction in order to discover potential issues. In addition, directly using ANN with posture prediction provides a preliminary step towards using ANN to predict the most appropriate combination of performance measures (PMs) - what drives human behavior. The PMs, which are the cost functions that are minimized in the posture prediction problem, are typically selected manually depending on the task. This is perhaps the most significant impediment when using posture prediction. How does the user know which PMs should be used? Neural networks provide tools for solving this problem. This thesis hypothesizes that the ANN can be trained to predict human motion quickly and accurately, to predict human posture (while considering external forces), and to determine the most appropriate combination of PM(s) for posture prediction. Such capabilities will in turn provide a new tool for studying human behavior. Based on initial experimentation, the general regression neural network (GRNN) was found to be the most effective type of ANN for DHM applications. A semi-automated methodology was developed to ease network construction, training and testing processes, and network parameters. This in turn facilitates use with DHM applications. With regards to motion prediction, use of ANN was successful. The results showed that the calculation time was reduced from 1 to 40 minutes, to a fraction of a second without reducing accuracy. With regards to posture prediction, ANN was again found to be effective. However, potential issues with certain motion-prediction tasks were discovered and shed light on necessary future development with ANNs. Finally, a decision engine was developed using GRNN for automatically selecting four human PMs, and was shown to be very effective. In order to train this new approach, a novel optimization formulation was used to extract PM weights from pre-existing motion-capture data. Eventually, this work will lead to automatically and realistically driving predictive DHMs in a general virtual environment.

Human driving decisions are the leading cause of road fatalities. Autonomous driving naturally eliminates such incompetent decisions and thus can improve traffic safety and efficiency. Deep reinforcement learning (DRL) has shown great potential in learning complex tasks. Recently, researchers investigated various DRL-based approaches for autonomous driving. However, exploiting multi-modal fusion to generate pixel-wise perception and motion prediction and then leveraging these predictions to train a latent DRL has not been targeted yet. Unlike other DRL algorithms, the latent DRL algorithm distinguishes representation learning from task learning, enhancing sampling efficiency for reinforcement learning. In addition, supplying the latent DRL algorithm with accurate perception and motion prediction simplifies the surrounding urban scenes, improving training and thus learning a better driving policy. To that end, this Ph.D. research initially develops LiCaNext, a novel real-time multi-modal fusion network to produce accurate joint perception and motion prediction at a pixel level. Our proposed approach relies merely on a LIDAR sensor, where its multi-modal input is composed of bird's-eye view (BEV), range view (RV), and range residual images. Further, this Ph.D. thesis proposes leveraging these predictions with another simple BEV image to train a sequential latent maximum entropy reinforcement learning (MaxEnt RL) algorithm. A sequential latent model is deployed to learn a more compact latent representation from high-dimensional inputs. Subsequently, the MaxEnt RL model trains on this latent space to learn a driving policy. The proposed LiCaNext is trained on the public nuScenes dataset. Results demonstrated that LiCaNext operates in real-time and performs better than the state-of-the-art in perception and motion prediction, especially for small and distant objects. Furthermore, simulation experiments are conducted on CARLA to evaluate the performance of our proposed approach that exploits LiCaNext predictions to train sequential latent MaxEnt RL algorithm. The simulated experiments manifest that our proposed approach learns a better driving policy outperforming other prevalent DRL-based algorithms. The learned driving policy achieves the objectives of safety, efficiency, and comfort. Experiments also reveal that the learned policy maintains its effectiveness under different environments and varying weather conditions.

A growing number of applications involving autonomous mobile robots will require their navigation across environments in which spaces are shared with humans. In those situations, the robot’s actions are socially acceptable if they reflect the behaviours that humans would generate in similar conditions. Therefore, the robot must perceive people in the environment and correctly react based on their actions and relevance to its mission. In order to give a push forward to human-robot interaction, the proposed research is focused on efficient robot motion algorithms, covering all the tasks needed in the whole process, such as obstacle detection, human motion tracking and prediction, socially aware navigation, etc. The final framework presented in this thesis is a robust and efficient solution enabling the robot to correctly understand the human intentions and consequently perform safe, legible, and socially compliant actions. The thesis retraces in its structure all the different steps of the framework through the presentation of the algorithms and models developed, and the experimental evaluations carried out both with simulations and on real robotic platforms, showing the performance obtained in real–time in complex scenarios, where the humans are present and play a prominent role in the robot decisions. The proposed implementations are all based on insightful combinations of traditional model-based techniques and machine learning algorithms, that are adequately fused to effectively solve the human-aware navigation. The specific synergy of the two methodology gives us greater flexibility and generalization than the navigation approaches proposed so far, while maintaining accuracy and reliability which are not always displayed by learning methods.

Properly accounting for the uncertainties in predicting ground motion parameters is critical for Probabilistic Seismic Hazard Analyses (PSHA). This is particularly important for critical facilities that are designed for long return period motions. Non-ergodic PSHA is a framework that allows for this proper accounting of uncertainties. This, in turn, allows for more informed decisions by designers, owners and regulating agencies. The ergodic assumption implies that the standard deviation applicable to a specific source-path-site combination is equal to the standard deviation estimated using a database with multiple source-path-site combinations. The removal of the ergodic assumption requires dense instrumental networks operating in seismically active zones so that a sufficient number of recordings are made. Only recently, with the advent of networks such as the Japanese KiK-net network has this become possible. This study contributes to the state of the art in earthquake engineering and engineering seismology in general and in non-ergodic seismic hazard analysis in particular. The study is divided in for parts. First, an automated protocol was developed and implemented to process a large database of strong ground motions for GMPE development. A comparison was conducted between the common records in the database processed within this study and other studies. The comparison showed the viability of using the automated algorithm to process strong ground motions. On the other hand, the automated algorithm resulted in narrower usable frequency bandwidths because of the strict criteria adopted for processing the data. Second, an approach to include path-specific attenuation rates in GMPEs was proposed. This approach was applied to a subset of the KiK-net database. The attenuation rates across regions that contains volcanoes was found to be higher than other regions which is in line with the observations of other researchers. Moreover, accounting for the path-specific attenuation rates reduced the aleatoric variability associated with predicting pseudo-spectral accelerations. Third, two GMPEs were developed for active crustal earthquakes in Japan. The two GMPEs followed the ergodic and site-specific formulations, respectively. Finally, a comprehensive residual analysis was conducted to find potential biases in the residuals and propose models to predict some components of variability as a function of some input parameters. / Ph. D.