Search

Networks such as transportation, water, and power are critical lifelines for society. Managers plan and execute interventions to guarantee the operational state of their networks under various circumstances, including after the occurrence of (natural) hazard events. Creating an intervention program demands knowing the probable direct and indirect consequences (i.e., risk) of the various hazard events that could occur in order to be able to mitigate their effects. This paper introduces a methodology to support network managers in the quantification of the risk related to their networks. The methodology is centered on the integration of the spatial and temporal attributes of the events that need to be modeled to estimate the risk. Furthermore, the methodology supports the inclusion of the uncertainty of these events and the propagation of these uncertainties throughout the risk modeling. The methodology is implemented through a modular simulation engine that supports the updating and swapping of models according to the needs of network managers. This work demonstrates the usefulness of the methodology and simulation engine through an application to estimate the potential impact of floods and mudflows on a road network located in Switzerland. The application includes the modeling of (i) multiple time-varying hazard events; (ii) their physical and functional effects on network objects (i.e., bridges and road sections); (iii) the functional interrelationships of the affected objects; (iv) the resulting probable consequences in terms of expected costs of restoration, cost of traffic changes, and duration of network disruption; and (v) the restoration of the network., Natural Hazards and Earth System Sciences, 18 (8), ISSN:1561-8633, ISSN:1684-9981

Complex engineered systems, such as infrastructures, power plants, and aircraft, are important for the proper functioning of modern societies and economies. Failures of these safety-critical systems can cause significant economic disruption and result in high social costs. In the interest of avoiding such failures and preserving the functionality of safety-critical systems in a cost-efficient manner, intelligent maintenance strategies are required. Concretely, maintenance interventions need to be planned according to the actual system health condition and its expected evolution over time rather than on the basis of predefined time intervals. Such intervention planning requires models that analyze the current and future system health condition in real time to provide early detection and isolation of fault types (i.e., diagnostics) and predict the system's expected failure time (i.e., prognostics). Over the last several decades, a large range of physics-based and data-driven models have been proposed to address fault diagnostics and prognostics problems. However, the applicability of these models to complex systems in real-world domains is often limited by two major challenges: (1) physical degradation processes are only well understood for critical or relatively simple components, resulting in incomplete physics-based models; (2) the datasets available for training data-driven models generally contain only a small number of fault types and are not, therefore, fully representative of all potential degraded system conditions. The combination of physics-based and data-driven models has attracted the research community's interest as a promising direction to overcome their individual limitations and obtain robust models for diagnostics and prognostics. As earlier works have pointed out, the combined use of physics-based and data-driven models can potentially increase their predictive performance on datasets with limited representativeness and compensate for the incomplete representation of the physics-based models. However, most of the earlier works combining physics-based and data-driven models do not take advantage of recent advances in deep learning, thus restricting their predictive performance. The more recent solutions using deep learning models are often restricted to fault types described with cumulative damage models and relatively simple physics-based models, often in the form of differential equations. This thesis aims to develop a framework and corresponding methods combining physics-based and data-driven models in order to provide accurate diagnostics and enable the prediction of the remaining useful lifetime of a fleet of complex systems. Concretely, this thesis proposes a novel framework for fusing the information from physics-based performance models with deep learning models for fault detection, diagnostics, and prognostics of complex safety-critical systems. In the proposed framework, physics-based performance models are used to infer unobservable model parameters related to the system components' health, solving a calibration problem. These parameters are subsequently combined with sensor readings and used as input to deep neural networks, thereby generating data-driven fault detection, isolation, and prognostics models with physics-augmented features. Besides the novelty in the information fusion of physics-based and data-driven models, several additional contributions are proposed for robust, accurate, and informative fault detection, diagnostics, and prognostics under challenging realistic scenarios: First, implicit supervision of healthy operative conditions is proposed in order to detect and distinguish different fault types (i.e., segmentation) with access only to healthy labeled data. Second, accurate detection and isolation of complex fault types affecting multiple sensors is demonstrated with deep learning-based method using physics-augmented features, which outperform alternative physics-based and data-driven models and overcome the limitations of the traditional multi-threshold-based methods. Third, the combination of physics-augmented features and deep learning is extended to the prognostics problem to provide an accurate estimation of the failure time in conditions of non-abundant failure data. The experimental results show that the proposed method outperforms purely data-driven models, requires less training data, and is less sensitive to the limited representativeness of the training dataset. Fourth, real-time and robust inference of unobservable model parameters related to a system's components' health is proposed with a reinforcement learning-based method and a tracking formulation of the calibration problem. Fifth, the proposed framework and methods were quantitatively validated on two synthetic datasets of turbofan engines under realistic operating conditions. Both datasets have been developed in the context of this thesis. They have been made openly available to facilitate the implementation of the proposed framework and facilitate further developments in prognostics and health management. In summary, the proposed framework and methods enable the detection, segmentation, and isolation of previously unseen fault types under realistic scenarios involving varying operating conditions and highly correlated signals. Moreover, they enable the prediction of the remaining useful lifetime in conditions of non-abundant run-to-failure degradation trajectories from a fleet of complex systems. Therefore, it is expected that the proposed framework can contribute significantly to the implementation of intelligent maintenance strategies for complex systems in the field.

Supply chain management plays a fundamental role in running a successful construction project to ensure cost-efficient, accurate, and on-time material delivery to construction sites. Supply chain management has been favored by researchers and construction practitioners over traditional material management practices that focus on a single logistic perspective. It can be used to manage interdependent project phases and improve stakeholder relationships. Although supply chain management helps stakeholders complete construction projects, the problems associated with material flow processes, such as material delays, rework, and incorrect deliveries remain frequent due to a lack of efficient supply chain integration and coordination. As a result, projects often face schedule delays and cost overruns. To keep construction projects on time and on budget, novel methodologies are needed for the management of material flow processes. Good management of material flow processes requires efficient coordination of material and associated information flows, which leads to reduced misalignment in deliveries. Responses to this need often focus on the use of advanced management principles, technologies, and intelligent algorithms to manage the supply chains. Lean management principles and Industry 4.0 technologies combined with various optimization algorithms have dominated manufacturing supply chains in the past decade and have the potential to enhance stakeholder collaboration and workflow efficiency. The supply chain processes in construction projects are normally dynamic in nature and are subjected to open-air environmental conditions. This adds considerable uncertainties to the management of material flows from the planning phase to the installation phase, in contrast to the manufacturing processes that happen in a closed and consistent work environment. Data silos further limit the capabilities of construction stakeholders to collaborate and inhibit the reliability of their decisions on material supplies and demands. To overcome the challenges and limitations, prevailing principles and methods in manufacturing industry are required to address the needs of construction supply chains. In pursuit of this, the dissertation aims to \textit{develop a systematic approach and the corresponding techniques to improve the coordination of material and information flows in construction projects, considering the integration of lean workflows, digital technologies, and optimization algorithms, with a special focus on stakeholders’ collaborative decision-making processes}. The first part of the dissertation focuses on the comprehensive review of contemporary supply chain management in construction projects. Various methods that facilitate construction supply chain coordination are identified through different review processes. One review is conducted to develop an objective and data-driven assessment of the use of automation and digital technologies to provide a better understanding of their potential benefits and limitations. The other review summarizes useful enablers to facilitate the coordination activities and to link different construction supply chain stages (i.e., design-to-production, production-to-logistics, and production-to-site-assembly). Results from both reviews suggest that a holistic integration of the supply chains and stakeholders’ collaborative decision making through digitalization (e.g., integrated BIM-IoT technologies) together with and lean workflows are essential to improve supply chain coordination but are missing from current practices. This part lays the knowledge foundation for the second and third parts of the dissertation. The second part of the dissertation explores the combination of lean principles and digital technologies; namely, lean workflows with integrated management systems. The lean workflow is designed to link digital information on material demands with look-ahead plans. This is illustrated in the material flow processes for the erection of an office building with prefabricated columns. The performance of the lean workflow is compared with that of a traditional workflow using discrete-event simulations. The lean workflow advances the traditional workflow as it allows project participants to combine detailed look-ahead plans with BIM and RFID technologies to better manage material flow processes. It is particularly useful for the management of engineer-to-order components considering dynamic site progress. To embed the lean workflow in a digital platform, an integrated management system is developed that focuses on the communication of design-change and schedule-change information. The system includes design-change, schedule-change, production, and transport functional modules. Project stakeholders can prevent material flows from the negative impact of late change using the system to digitally support the lean workflow. The system uses a client-server architecture, though it can be easily reconfigured to a cloud-based system architecture. The capability of the system to improve the coordination of changes over an engineer-to-order material flow process simulation of the office building project is demonstrated to be superior than that of the traditional management framework. This work represents the first instance where detailed look-ahead plans, BIM, RFID, and a common data environment are integrated in supply chain collaboration to replace otherwise ad-hoc procedures. The third part of the dissertation moves the focus to the management of bulk commodity. This part explores the combination of lean principles, digital technologies, and optimization algorithms to improve stakeholders’ decision-making processes in the coordination of material flows. This part selects ready-mix concrete as a representative type of bulk commodity to study this dynamic. Given the fluctuating demand for concrete, an approach that uses digitized information within a short-term planning window is presented to improve the ordering of concrete by minimizing waste and cost. The approach consists of two main steps that contractors should take: 1) monitoring the dynamic demand fluctuations using a 4D model that captures the as-built site progress and updated look-ahead schedules, 2) modifying original orders to accommodate the demand fluctuations where quantities of additional orders and outsourced orders are determined by a heuristic evolutionary algorithm. The proposed approach is demonstrated to be useful to provide optimal order quantities of concrete with minimal costs within a five-day planning window. The scope of implementing the approach is expanded from a single project to a group of projects in a contractor’s or owner’s broader portfolio. To optimize the allocation of materials among multiple projects, a transshipment method is developed to enable the lateral sharing of materials in a supply chain network using the same two-step approach; a streamlined coordination process comprising the extraction of digital information and the decision-making on material allocation. To accommodate the expansion of the approach to varied material demands among projects, a network model is required to simulate the material and information flows from one project to another. Either within or among projects, the implementation of the two-step approach in both cases proves the benefits of balancing the daily material supply and demand according to a continuously updated schedule. As a result, the materials can be managed with minimal waste and costs. This dissertation contributes to the field of construction management of material flow processes by providing a systematic approach and corresponding techniques to address the challenges in construction supply chain processes. More precisely, it provides a novel method for integrating lean principles, digital technologies, and optimization algorithms for construction supply chain practices. The proposed approach improves supply chain efficiency, transparency, agility, sustainability, and accuracy, paving the way for a better future of construction projects., ISBN:978-3-907234-30-3

Modern societies are heavily reliant on the service provided by water distribution networks. Despite the critical importance of this infrastructure, water utilities are facing major financial challenges worldwide and are raising water fees in response. The willingness of stakeholders to bear this growing financial burden is low. Without clear and convincing argumentation, future water fee hikes will become increasingly difficult to obtain. The infrastructure asset management process is a well-suited tool for providing such argumentation. This process is a systematic approach to ensure that infrastructure continues to provide an adequate level of service during all lifecycle phases (i.e. construction, operation, maintenance, expansion, and demolition). Over the last several decades, digitalization has permeated all activities related to infrastructure asset management and led to tangible benefits such as the improved design, efficient operation and timely maintenance of water distribution networks. However, compared to other sectors of the economy, digitalization in the infrastructure sector is lagging. Concrete examples are needed to nudge infrastructure managers towards further embracing digitalization in all aspects of the infrastructure asset management process. This thesis aims to fulfil this need by providing specific examples of how digitalization dissolves data silos within utilities and between them and enables the adoption of efficient algorithms to tangibly improve asset management of water distribution networks. The contents of the thesis is divided into three main parts. The main themes of the three parts are pipe failure prediction with artificial neural networks (ANNs), optimal intervention planning for water distribution networks, and optimizing the execution of operational maintenance programs. Part I consists of Chapter 2. Part II consists of Chapters 3-6. Part III consists of Chapter 7. This work contributes to the advancement of infrastructure asset management of water distribution networks in several ways. Firstly, a study examines how a pipe failure prediction methodology based on ANNs can be used to make short-term failure predictions on all pipe materials by combining soft deterioration data from an expert elicited survey with hard deterioration data from recorded failures. Secondly, a methodology is presented for determining optimal short-term intervention programs on large, real-world water distribution networks that considers all network components requiring preventive intervention planning. Several aspects of the novel methodology are explored in additional chapters. These include the use of cost-benefit analysis in quantitatively comparing interventions on different object types, the influence of economies of scale on pipe replacement estimates, and the simplifications used to model large, real-world networks. Finally, a study is provided exploring the influence of maintenance density and routing algorithm on the efficiency gains that can be achieved by integrating municipal maintenance tasks such as hydrant and valve inspections and meter replacements into combined work packages. The most substantial contribution of this thesis is found in Part II on optimal intervention planning. The increasing demands of stakeholders for decision transparency, resource efficiency and sustainability will continue to serve as catalysts for the development and professionalization of infrastructure asset management. The untapped potential of digitalization represents a clear path forward to achieving these ambitious goals. Ultimately, digitalization will not only transform decision-making but will also fundamentally reshape the organizational structures within utilities as well as the water sector itself., ISBN:978-3-907234-37-2

Infrastructure systems are important for social development and economic growth. They play a fundamental role in the use and distribution of spatial services, such as transportation and communication. Recent historical events such as the 2002 European flood, hurricane Katrina (2005), or the Tōhoku earthquake and tsunami (2011) have shown that the analysis and understanding of large-scale infrastructure systems are essential for research, engineering and society. Especially due to the complexity and interdependence of these systems, localised failures can cascade dramatically, leading to widespread, unforeseen and often disproportionate disruptions compared to the actual physical damage. Infrastructure managers plan and execute interventions to guarantee the operational state of their infrastructures under various circumstances. This also applies in the aftermath of natural hazard events. As the resources available to managers to protect their infrastructures are limited, it is essential for them to be aware of the probable consequences (i.e., risk) in order to set priorities and be resource-efficient. In order to support infrastructure managers in their risk assessments, this work aims to develop a methodology and corresponding techniques to understand and quantify the risk of complex infrastructure systems affected by natural hazards, considering spatial and temporal aspects. The first part of this work focuses on the development of a risk assessment process for infrastructure systems affected by natural hazards using computational models to simulate different hazard scenarios and estimate the associated consequences. In this part, a general risk assessment process for infrastructure systems affected by natural hazards is introduced. Based on this process a simulation engine is presented which is constructed as a computational platform to estimate risk by supporting the combination of models from different disciplines. This allows the application of the proposed process to estimate the spatio-temporal risk of a realistic road network due to the occurrence of time-varying multi-hazard events, considering physical and functional effects on network objects (e.g. bridges and road sections), the functional interrelationships of the affected objects, the resulting probable consequences, duration of network disruption, and the restoration of the network. To give better insights into the resilience of the infrastructure system to natural hazards and to help the infrastructure managers to make better decisions in such situations, a restoration model is formulated to determine optimal recovery responses in the aftermath of such events. While the implementation of the risk assessment process in the first part is mainly based on computational models, the second part focuses on the development of innovative mathematical models from the field of network sciences. First, a network model for interdependent infrastructure systems is presented, which is based on the mathematical concept of the spatially embedded random network, and therefore, needs only a limited amount of data. Second, a complex network approach is used to investigate traffic flow dynamics on road networks, which provides reasonable estimates for traffic flow changes and significantly reduces the computing time of classical simulation models. These models can be used to support the risk assessment of complex infrastructure systems in terms of data requirements and computing power, i.e. computational models can be substituted if only limited data is available, or a decrease in the computational effort is required. This work contributes to the field of risk assessment and its application to complex infrastructure systems, by providing a methodology and corresponding techniques to understand and quantify the risk of complex infrastructure systems, affected by natural hazards, considering spatial and temporal aspects. More precisely, this work provides a novel risk assessment process for infrastructure managers, designed to estimate the spatio-temporal risk of complex infrastructure systems due to the occurrence of time-varying multi-hazard events. In doing so, it not only extends the state-of-the-art research in this field but also helps to provide decision support for infrastructure managers., ISBN:978-3-906916-63-7

ISBN:978-3-906916-60-6

In this paper, a Cost-Performance-Indicator (CPI) curve and the respective CPI method are proposed to indicate additional costs and benefits of energy efficiency measures at a strategic level, and evidence is given that corroborates the hypothesis that energy efficiency measures follow the law of increasing relative costs. The usefulness of the CPI curve is demonstrated through two case studies. An example is provided and the potential is discussed for using this curve for the planning and budgeting of refurbishment and energy efficiency measures, and as a tool to explain the relation between costs and benefits of measures enhancing building energy efficiency, including the production of renewable energy, to investors.

ISBN:978-3-906327-04-4