What I Do

Cyber-Physical Systems (CPS) are combinations of physical processes and embedded computers, which shape the basis of our future smart systems. The potential impact of these systems will far exceed the simple embedded systems of today. The use of CPS ranges from critical infrastructures such as smart cities, power and water grids, traffic networks and the Internet of Things (IoT) to smaller-scale personal and family items, such as smart phones, vehicles and wearable technologies. This notable integration of CPS give rise to many new challenges in reliability, safety, security, scalability and adaptability of these systems. My research contributes to the creation of expressive models of cyber-physical systems and the environment they are interacting with. I exploit a unique methodology to close the gap between system identification and machine learning for modeling these complicated systems and their interactions. The obtained models then are leveraged to develop scalable model-based techniques, which ensure reliability, safety and security of CPS. In addition, these techniques are capable of inferring the intention of humans or other interacting smart devices to enhance robustness and resilience of CPS. This is done by utilizing recently developed optimization tools such as mixed-integer programming and the relaxation tools available for polynomial optimization problems. The ultimate goal is then to design decision making mechanisms that are robust to uncertainty, resilient to faults and attacks, and tractable to implement in real-time.

In addition, I am actively involved in the global effort to integrate sustainable green resources into our future energy grids. Currently, the renewable sources benefit from subsidies introduced by government, which reduces competition, efficiency and social welfare. Alternatively, my research contributes to lowering the cost of renewable resources and other sustainable opportunities such as demand response (dynamic participation of consumers in grid operation) by mitigating the uncertainty and increasing the efficiency of these resources.

Intention Identification in Human-Robot Applications (LCSS*’17)

alt text 

The transition to autonomy in passenger and commercial vehicles is happening at an incredible pace. One of the major challenges for autonomous driving is to safely identify and react to the behavior of human-driven cars and other autonomous vehicles. In the present study, we employ a model-based optimization method to identify the intention of other drivers at intersections by taking appropriate control actions while simultaneously optimizing for safety, comfort and energy efficiency. In this work we developed an optimization-based Discriminating Input Design (DID) approach to guarantee identification of the intention of other cars at an intersection, while optimizing for efficiency, safety or comfort. This methodology is applicable to other human-robot interactions such as other driving scenarios, driver drowsiness detection and elderly care robots.

Guaranteed Passive/Active Fault and Attack Detection (Automatica’17, ADHS’15, ACC’16, IFAC WC’17)

Guaranteed Passive Fault Detection

alt text 

T-detectability: T-detectability is a property of a pair of system and fault models. If this property is satisfied for a given pair, and for a given T (an integer), it means that the behaviors of the two models are separable in T time steps. This allows us to implement fault detection or model invalidation methods on a receding horizon with a fixed size without compromising the detection guarantees. If a pair of system and fault models are T-detectable, then our guaranteed fault detection method can be implemented in real-time for wide variety of applications.

alt text 

In this work, we present a sound and complete fault detection approach for cyber-physical systems represented by hidden-mode switched affine models with time varying parametric uncertainty. The fault detection approach builds upon techniques from model invalidation. In particular, a set-membership approach is taken where the noisy input-output data is compared to the set of behaviors of a nominal model. As we show, this set-membership check can be reduced to the feasibility of a mixed- integer linear programming (MILP) problem, which can be solved efficiently by leveraging the state-of-the-art MILP solvers. In the second part of the paper, given a system model and a fault model, the concept of T-detectability is introduced. If a pair of system and fault models satisfies T-detectability property for a finite T, this allows the model invalidation algorithm to be implemented in a receding horizon manner, without compromising detection guarantees. In addition, the concept of weak-detectability is introduced which extends the proposed approach to a more expressive class of fault models that capture language constraints on the mode sequences. Finally, the efficiency of the approach is illustrated with numerical examples motivated by smart building radiant systems.

Guaranteed Passive Fault Isolation

alt text 

This research considers the problem of fault detection and isolation (FDI) for switched affine models. We first study the model invalidation problem and its application to guaranteed fault detection. Novel optimization-based formulations are proposed for model invalidation and T-detectability problems, which are both more intuitive and computationally efficient, without the need for complicated change of variables. Moreover, we introduce a distinguishability index as a measure of separation between the system and fault models, which offers a practical method for finding the smallest receding time horizon that is required for fault detection, and for finding potential design recommendations for ensuring T-detectability. Then, we extend our fault detection guarantees to the problem of fault isolation with multiple fault models, i.e., the identification of the type and location of faults, by introducing the concept of I-isolability. An efficient way to implement the FDI scheme is also proposed, whose run-time does not grow with the number of fault models that are considered. Finally, the effectiveness of the proposed method is illustrated using an HVAC system model with multiple faults.

Guaranteed Active Fault/Attack Diagnosis

alt text 

In this work, we consider the active model discrimination problem amongst a finite number of affine models with uncontrolled and noise inputs, each representing a different system operating mode that corresponds to a fault type or an attack strategy, or to an unobserved intent of another robot, etc. The active model discrimination problem aims to find optimal separating inputs that guarantee that the outputs of all the affine models cannot be identical over a finite horizon. This will enable a system operator to detect and uniquely identify potential faults or attacks, despite the presence of noise in the system process and measurements. Since the resulting model discrimination problem is a nonlinear non-convex mixed-integer program, we propose to solve this in a computationally tractable manner, albeit only approximately, by proposing a sequence of restrictions that guarantees that the obtained input is separating. Finally, we apply our approach to attack detection in the area of cyber-physical systems security.

Integration of Sustainable Resources into Future Energy Grids (VPPC’07,SmartGridComm’14,CDC’16)

The significant increase in the percentage of greenhouse gases in the atmosphere since the beginning of the industrial revolution has led to global warming and other major environmental challenges. This is mainly due to the utilization of fossil fuels as energy sources. As such, transformation from fossil fuel energy to sustainable renewable sources is inevitable. The problem is, however, the uncertainty associated with renewables. The focus of the second direction of my research is to offer solutions to mitigate the uncertainty and lay foundation for large penetration of renewable energy sources in future smart grids. In addition, I will pursue research to take full advantage of green sustainable opportunities in demand response to provide reserve and ancillary services to the electricity markets. My interests encompass (1) the economical aspects, i.e., designing incentive cost sharing mechanisms for sustainable energy markets and (2) developing novel and computationally tractable scheduling, estimation and control techniques for aggregations of supply and load to mitigate uncertainty and reduce the costs.

Optimal Payment Sharing Mechanism to Entice Aggregation of Renewable Suppliers

alt text 

The participation of renewable energy sources in energy markets is challenging, mainly because of the uncertainty associated with them. Aggregation of renewable energy suppliers is shown to be very effective in decreasing this uncertainty. In this work we propose a new mechanism to share the profit of the aggregate amongst individual producers. Calculating optimal contract for individual producers and the aggregate is the first requirement in order to propose a payment sharing mechanism that has been addressed in the literature. Different researchers utilized various methods from coalitional game theory to statistical methods to introduce payment sharing mechanisms. In this work, we propose a novel payment sharing mechanism that maximizes the aggregate profit in first place. The payments to the individuals are based on their actual production instead of the expected. We also show that this method will pay more to the individuals in average compared to the previous methods.

Effect of Bonus in Payment Sharing Mechanisms

alt text 

In this work, we propose a cost sharing mechanism that entices the suppliers of wind, solar and other renewable resources to form or join an aggregate. In particular, we consider the effect of a bonus for surplus in supply, which is neglected in previous work. We introduce a specific proportional cost sharing mechanism, which satisfies the desired properties of such mechanisms that are introduced in the literature, e.g., budget balancedness, ex-post individual rationality and fairness. In addition, we show that the proposed mechanism results in a stable market outcome. Finally, the results of the paper are illustrated by numerical examples.

Smart Manufacturing (CDC’13,IEEE TSM’14a,IEEE TSM’14b)

alt text 

Run-to-run control is used in the semiconductor manufacturing industry to stabilize unit processes within the manufacturing sequence. The concept of run to run control is illustrated in figure. A wafer arrives at a particular process step, and is processed using settings determined by a controller. In run to run control, these settings do not depend on in-situ measurements of the process, because of the size of metrology devices, thus they are completely determined at the time of wafer arrival at the process step. Later, this process is measured ex-situ to determine the results of the processing step. These measurements are given to the run to run controller, which will use them to determine the appropriate process settings for future wafers.

Thread Initialization in Run-to-Run Control of Lithography Process

alt text 

Run to run control is a major tool used in semi- conductor manufacturing to keep the unit processes within the required manufacturing constraints. Typically, the difference, or bias, between the desired and actual result of processing a particular wafer is affected by not only the particular product being produced, but the prior processing path. Each unique combination of effects is called a thread (shown in figure), and in threaded run-to-run control a separate EWMA controller is applied for each thread. One of the challenges of threaded control is the initialization of the bias estimate for a new thread. Automated initialization methods prevent the cost of manual initialization or utilizing a pilot run, but at the risk of producing outliers in the initialized runs. On the other hand the manual initialization or using pilot runs incur an extra expense. In this work we study a new approach for initialization of the threads by combining the threaded and non-threaded control strategies. This method avoids the cost associated with manual initialization methods and improves the efficiency of automated methods. Finally the simulation results will demonstrate the efficiency of the proposed method in comparison with other techniques.

Resolving Structural and Data Sequence Based Unobservability in Non-Threaded Run-to-Run Control

alt text 

In the literature, non-threaded run-to-run control methods have been presented which describe the process bias for a particular wafer as a linear combination of possible bias contributions, with the individual contributions estimated using a Kalman Filter. In this paper, we address two issues that need to be considered in implementations: observability of the state realization of the bias model, and the computational cost of the Kalman filter. While some elements of the bias model that create unobservability are well known, we present a complete analysis of observability that also considers the influence of the thread sequence. We also survey and extend methods of observability recovery that do not require model reduction or the specification of special reference threads, thus easily allowing new threads to be added and old threads removed. Finally, we describe how the problem structure allows the information form of the Kalman filter to be much more computationally efficient. Simulation results illustrate the proposed method.