Pre-Conference Workshops / Tutorials
ACODS 2020 offers four exciting full day (9 AM to 5 PM) pre-conference workshops / tutorials by leading researchers in the respective areas:
- Overview and classification of online process optimization approaches
Workshop organizers: Dinesh Krishnamurthy and Sigurd Skogestad, NTNU, Norway Summary: This workshop discusses various approaches to online process optimization, where the objective is usually to minimize an economic cost. Steady-state real-time optimization (RTO) has been around for more than 25 years, but still it is not used much in practice. One reason is the steady-state wait time, because one has to wait for a new steady state before the process is re- optimized. In this workshop, a number of alternative approaches are discussed:
1. Traditional steady-state RTO
2. Dynamic real-time optimization (DRTO) / Economic (nonlinear) model predictive control (EMPC)
3. Hybrid RTO – Steady-state RTO with dynamic model update (new method)
4. Feedback-based Hybrid RTO (new method)
5. Extremum seeking control (“Data-driven” optimization)
6. Self-optimizing control
7. Optimal operation using classical advanced control
8. Machine learning tools for process optimizationExcept possibly for EMPC, the optimizer sends setpoints to a control layer, which could be a PID layer or MPC. The lower control layer then handles dynamic stability issues. In the conventional steady state RTO, one must wait for the process to reach steady state before updating the model using “data reconciliation”. As a result, large chunks of transient data are often discarded. In order to address this issue, in the hybrid RTO approach, we solve the same steady-state optimization problem as in traditional steady state RTO, but instead of a steady-state model update, we use dynamic model adaptation with use of transient measurements, for example, using an extended Kalman Filter. This avoids the steady-state wait time. In the feedback-based hybrid RTO approach, we do not solve the steady-state optimization problem numerically as in conventional RTO, but instead the steady-state gradient is estimated by linearizing the nonlinear dynamic model around the current operating point. The gradient is controlled to zero using standard feedback controllers, for example, a PI-controller. Unlike model-based methods, data driven methods such as extremum seeking control and modifier adaptation that rely on estimating the plant-gradients directly from the measurements, can effectively handle structural mismatch. However, this requires the assumption that the plant cost can be measured and in addition, the convergence of such methods are generally very slow, due to the steady-state wait time. We show how transient measurements can be used along with such methods. We also present solutions using classical advanced control, where optimal operation can be achieved using PID controllers with simple logics such as split range control, selectors etc. Finally, with the recent surge of interest in machine learning and AI, we explore and discuss some research ideas on using such tools for addressing the challenges with real time optimization.
With the recent developments of various approaches to online process optimization with varying degrees of complexity and flexibility, different methods work in different timescales and can handle different kinds of uncertainty. This workshop will give an overview and classification of the different approaches available in the RTO “toolbox” and discusses the advantages and disadvantages of the different methods.
- Global Navigation Satellite Systems
Workshop organizer: Sanat Biswas, IIIT Delhi, India Summary: Advent of Global Positioning System (GPS)/Global Navigation Satellite System (GNSS) technologies have radically changed the navigation process. This tutorial aims to make the audiences familiar with space-based navigation, GPS/GNSS, GPS signal structure, acquisition and position computation techniques from GPS data. This tutorial also focuses on the fusion of GNSS and other available sensor observations to estimate accurate position and velocity, which is widely used for navigation. The tutorial will start with an introduction to radio navigation and GPS. After discussing GPS signal structure and positioning algorithms, various position error sources will be introduced, and various error correction techniques will be explained. GPS signal acquisition and receiver structure are also within the scope of this tutorial. Differential GPS and Space-Based Augmentation System (SBAS) will also be discussed. New GPS signals and other global and regional satellite navigation systems will be also introduced to the audiences.
Inertial Measurement Unit (IMU) will be introduced in the second part of the tutorial. The Kalman Filter and its non-linear variants will be discussed. The usage of GNSS and IMU observations in Kalman Filter framework to estimate accurate position and velocity will be explained at the end of the tutorial.
Postgraduate researchers, academics and practising engineers in the field of navigation, guidance and control will find this tutorial useful. Prior knowledge of signal processing, linear algebra, stochastic process is desirable.
- Nonlinear Real-time Optimal Control using MPSP: A New Fast MPC Paradigm
Workshop organizer: Radhakant Padhi, IISc, Bangalore, India Summary: Even though many challenging real-life problems can be formulated in the framework of nonlinear optimal control theory, it is well-known that it often gets trapped in the computational requirements. The classical calculus of variations approach often leads to a two-point boundary value problem formulation and lands up in the “curse-of- complexity” issue, because of which it is not a viable tool for online applications. Alternately, attempt to get state- feedback design following the philosophy of dynamic programming leads to the Hamilton-Jacobi-Bellman equation and gets trapped in the “curse-of-dimensionality” issue, and hence is not a viable tool for online applications either. Yet another approach, the transcription approach leads to large-dimensional optimization problems and gets trapped in slow convergence, local optimum etc. Because of these, despite having nice features, utility of nonlinear optimal control formulations has been primarily confined only to off-line applications (e.g. trajectory optimization). However, various attempts are being made in the recent past to overcome this computational complexity issue, thereby making the optimal control design approach a viable tool for online applications.
One such promising technique happens to be the recently-developed Model Predictive Static Programming (MPSP), which broadly falls under the fast model predictive control (MPC) paradigm. The main objective of this workshop will be to expose and equip the participants with the MPSP technique and its several extensions. Like model predictive control (MPC), a model-based prediction-correction approach is adopted in MPSP. However, the entire problem is converted to a very low-dimensional static programming problem, from which the control history update is computed in closed-form (i.e. without the need of an optimization solver) for most of the problems. Moreover, the necessary sensitivity matrices (which are the backbone of the algorithm) are computed recursively. These two key innovations make the process computationally quite efficient, thereby making it suitable for implementation in real- time. Note that MPSP proposes a ‘remarkable departure’ from classical optimal control approaches by updating the entire control history using a low-dimensional ‘static’ costate vector (i.e. just a Lagrange multiplier!). Other good feature happens to be rapid convergence of the update cycle and no-requirement of a good control history guess to initiate the process. Details of this very promising MPSP algorithm, along with several recent extensions, which make it applicable for a wide class of problems, will be discussed in this workshop. Several challenging real-life problems from aerospace, mobile robotics and process control will also be discussed in fair detail to demonstrate the generality and usefulness of the MPSP technique.
No prior knowledge on optimal control and/or MPC is necessary to attend this workshop. A good review of the basics of static optimization and optimal control theory will be done at the beginning to provide enough necessary background. A generic MATLAB function about the basic MPSP technique developed by the instructor will be distributed to the participants to enable them getting started quickly.
- Systems Biological Modeling and Algorithms for Management of Diabetes
Workshop organizer: Pramod Somvanshi and Sunil Deshpande, Harvard John A. Paulson School of Engineering and Applied Sciences, Cambridge, USASummary: Diabetes is a metabolic disorder characterized by blood glucose dysregulation which leads to poor health outcomes. The management of blood glucose regulation in diabetes is multifaceted requiring choice of diet, lifestyle and medications. The aim of this half-day session is to provide tutorial on the state-of-the-art results in systems biological modeling and algorithms for management of diabetes.
The session will be organized along two main topics: systems modeling and decision algorithms.
- Survey and discussion of mathematical models of metabolism in nondiabetic and
- Minimal model of insulin, glucagon and glucose homeostasis
- Integrative analysis of multiscale models for influence of plasma macronutrient levels
on homeostasis and metabolic disease states
- Data-driven system identification of metabolic processes for improved decision
- Survey and discussion of mathematical models of metabolism in nondiabetic and
- Design challenges in sensing, actuation and algorithms for closed-loop insulin delivery in people with type 1 diabetes
- Survey and discussion of algorithms for automated insulin and glucagon delivery used in clinical studies
With special theme of the conference on “Systems and Data Sciences for Engineering and Health Care”, this tutorial session will provide an opportunity for the audience of ACODS 2020, in particular graduate students and postdocs, to gain knowledge about both theoretical and practical aspects of this interesting problem in systems biology and biomedical control.
Please note the following guidelines before registering for these workshops:
- Register for these pre-conference tutorials via the same Papercept site that has been set up for conference registration.
- The registration fee is 1000 INR / 15 USD (all taxes inclusive) for students and 2000 INR / 25 USD (all taxes inclusive) for professionals.
- A full waiver for students / project associates is possible under special financial circumstances and on a case-by-case basis only. Requests for waiver should be sent as a document by email to firstname.lastname@example.org. Please note that the request should be signed by the student and endorsed by the research / project advisor. The conference organiser’s decision will be notified to the concerned student by email.
- In the event that a workshop is cancelled for lack of insufficient registration or for any other unforeseen reason, the registration fee for the workshop shall be refunded to the participant.