Stochastic optimization and simulation
The term "stochastic" originates from the Greek word meaning "able to guess" and, in modern times, specifically refers to modeling uncertain phenomena using a probability measure. The future behavior of most real-life operational systems—whether in industry or elsewhere—is influenced by parameters that carry some degree of uncertainty. For example, how long a warehouse operator takes to pick a customer's order, the time a vehicle needs to travel between addresses, the impact of weather on a solar panel's output, a company's sales figures for the next quarter, or a patient's consultation time are all uncertain factors that can be effectively modeled as random variables with specific probability distributions. This distribution can be suggested by the modeler based on domain expertise or empirically derived from historical data. Naturally, when some input parameters of a system's model are random variables, so too are the system's resulting performance metrics—such as the number of orders picked in a warehouse per day, the fuel cost of a vehicle's delivery route, today's energy costs for a production plant, a company's quarterly profit, or a patient's waiting time. To draw meaningful insights, we aim to compute or estimate the expected value of these random performance metrics.
For many realistic systems, however, analytically evaluating the desired performance metrics is often unfeasible without resorting to crude approximation methods. Fortunately, with sufficient computational power, we can reliably estimate them through sampling techniques such as Monte Carlo simulation. Although conceptually straightforward, the main challenge here is to build a computer model of the system's functionality and evolution over time, striking a balance between simplicity and complexity relative to the performance metrics we aim to estimate. Discrete-event simulation is one of the well-established techniques for achieving this balance effectively.
Stochastic optimization builds on this idea. Suppose that, in addition to some uncertain parameters, the system also includes decision variables under the control of decision-makers. Examples include the number of picking operators in a warehouse, the size of the delivery vehicle, the number of solar panels installed, the production quantity in a quarter, or the number of doctors scheduled to receive patients. If we can evaluate or estimate the (expected) performance metrics of the system and translate them into financial costs, then stochastic optimization seeks to find the optimal values of the decision variables to minimize those costs. The structure of the problem determines how efficiently this can be achieved, but in most cases, significant computational resources are required, and even then, it's rare to guarantee that the best-found decision is truly optimal. The primary challenge in stochastic optimization is often to strike a balance between the quality of the obtained decision and the computational time required to find it.
Industrial planning of a continuous reactor powered by renewable energy
Powering reactors with high energy demands becomes more sustainable when renewable energy is employed. Accounting for the intermittent nature of renewable energy when planning the energy flow improves the efficiency of the system. Stochastic models are used to account for forecasting errors in energy production and energy prices.
Intelligent assembly methods to enable sustainable manufacturing
Efficient use of raw materials is essential for sustainable manufacturing. Hybrid and selective assembly are effective ways to relax tolerances and reduce scrap without affecting the final product's quality, thereby optimizing resource usage. We evaluate their operational and economical implications to support decision-makers in adopting intelligent assembly strategies.
Carbon-aware production scheduling
Growing awareness of environmental concerns and increasingly stringent regulations are driving industries to reduce the carbon footprint of their operations. By incorporating dynamic carbon content of grid energy and on-site renewables, carbon-aware scheduling can serve as a powerful way to reduce manufacturing's carbon footprint. We develop models to minimize scope 2 emissions, along with metaheuristics for efficient solutions.
Data-driven optimization for planning and scheduling in flexible manufacturing systems
In a flexible manufacturing system (FMS), scheduling stands as a major component that plays a critical role in ensuring the efficiency and effectiveness of the system. Scheduling in an FMS is highly complex due to multiple factors, including the presence of multiple resources such as machines and operators, the multiplicity of routings for the processing of jobs, the existence of multi-functional machines, and the inherent uncertainty of some parameters, such as task durations and machines’ reliability. The goal of the research carried out is essentially to develop high-performing optimization solution approaches for the integrated flexible job shop and operator scheduling, while taking uncertainty of critical parameters, such as task durations, into account.
Well-being-aware task allocation and scheduling in complex manufacturing systems
Traditionally, tasks are assigned to operators and scheduled according to classical inputs, such as the sets and features of tasks, operators and machines, and constraints between them. However, this approach often results in some operators bearing disproportionately high ergonomic and cognitive loads, while others experience little to no pressure. Therefore, we develop task allocation and scheduling algorithms that also integrate human factors. These algorithms should contribute to significantly mitigate the high absenteeism rates among operators. Notably, in the 2014/15 period, Work-Related Musculoskeletal Disorders (WRMSDs) accounted for 9.5 million lost working days in the UK alone.
Integrated tactical production planning for factories involved in strategic industrial symbiosis collaborations
We consider a supply chain in which the involved factories are engaged in a strategic industrial symbiosis collaboration. An industrial symbiosis occurs when an industrial activity generates waste or by-products that become a resource for another industrial activity. To take full advantage of this industrial symbiosis collaboration, it is necessary to provide the decision-makers with a framework to deploy this collaboration at all levels of the planning. The research carried out focuses on integrated tactical production plans of the factories engaged in this industrial symbiosis collaboration, while considering uncertainty related to their corresponding markets. Different approaches, including stochastic and robust optimization, are used to produce well performing integrated plans in terms of solution quality and computational time.
Decision support models for part feeding in flexible assembly lines
With the rising demand for customized products, flexibility in mixed model assembly lines becomes more critical. Even so, as the variety of products in assembly lines grows, the number of parts required at the border of line (BoL) also increases. To effectively utilize the limited space at the BoL, appropriate line feeding policies must be implemented. In this case however, additional costs related to preparation and transportation will be incurred. The goal of this research is to develop robust optimization methods for tactical assembly line feeding decisions, aiming to minimize feeding costs while maintaining flexibility in mixed model assembly lines.
Spare parts inventory control in a condition based maintenance context
The Challenge: For manufacturers of capital equipment, after-sales service is the new battleground for customer loyalty and profitability. But managing spare parts efficiently can be a costly headache.
The Solution: Integrating condition monitoring information into spare parts inventory decision making. We develop algorithms that leverage this information to optimize inventory control, ensuring parts are available when needed, while maintaining low inventory levels.
Stochastic optimization models for allocation of quality inspection in flexible assembly
Inspection in a complex assembly environment can be done using many types of tests, sensors and technologies, each with their own performance (in terms of precision, speed, sensitivity, specificity), resource requirements and involved operating costs. We also need to decide at which positions of the assembly inspection will be performed, which characteristics of the assembly will be inspected and what inspection strategy to use (full inspection, sampling, ...). It is a challenge to find an efficient allocation of the inspection effort so that the end product conforms to customer specifications at minimal cost. The PhD aims to develop innovative cost optimization models that take into account detailed knowledge of the assembly process such cost of rework, parts and operator time, specifications, warranty cost, bill of materials, but also knowledge of different eligible tests. The outcome is a data driven decision support system for optimal inspection allocation in the assembly environment. It will also focus on the robustness of the solutions obtained by the model to (unforeseen) changes in the production schedule. This is highly relevant in high-mix low-volume assembly where production runs are short and change frequently in nature. The inspection solutions must be flexible enough to deal with this variability.