At CPSL, our overarching goal is to explore and understand the synergies across large scale sensor-driven inference and decision making. We advance the integration of mathematical programming and data analytics in various application domains. Specifically, we focus on the modeling and the computational challenges arising from the integration of real-time inferences generated by advanced data analytics and simulation into large-scale mathematical programming models used for optimizing and controlling networked systems.
What is unique about our approach? There is a growing body of literature and commercial applications on sensor driven failure prediction methods that estimate the current state of health (i.e. diagnostics), and the remaining life distribution (i.e. prognostics) of cyber-physical assets. It is often difficult to translate these failure predictions to operations and maintenance decisions. State-of-the-art methods schedule immediate maintenance actions for assets with imminent failure risks. These immediate repairs often impose significant toll for operational performance. In reality, failure predictions can be harnessed for proactive decision-making models that explicitly capture and jointly optimize the interdependencies across assets, operations and maintenance in large scale networks. Our approach offers a seamless integration of mathematical programming and data analytics to bridge the gap across predictive and prescriptive models.
What is the potential industrial impact? Operations & maintenance (O&M) outcomes are central concerns for cyber-physical systems with far-reaching implications for profitability, reliability & resilience. At an asset-level, O&M costs constitute a significant contributor to costs (e.g. >50% of running costs in marine energy, >17% of LCOE in solar energy). At a system level, sudden failure events impose significant risks to network operations, & cause wide-spread generation unavailability. O&M also creates a significant barrier for competitiveness of the emerging renewable energy technologies, e.g. tidal and wave energy systems.
What are the associated technical challenges? Fundamental research challenges in modeling and computation arise from the integration of real-time sensor-driven inferences into large-scale mixed integer programs (MIPs) used for optimizing networked energy systems and markets. On the modeling side, I offer novel reformulation approaches to embed continuous-time continuous-state degradation functions within MIP models as sets of constraints and uncertainty models. Significant impact of operational loading on asset degradation is inherently modeled via decision-dependent uncertainty. The resulting optimization models are augmented with case-specific operational considerations (e.g. unit commitment, wind farm operations). On the computation side, I propose new optimal solution algorithms that exploit the structure of the optimization models to enhance computational scalability. My team in CPSL brings together expertise from electrical, mechanical & industrial engineering, and computer science to offer a novel sensor-driven lens to some of the most fundamental operations and maintenance problems.
What are the associated benefits? Our research approach offers the following benefits: