National Science Centre (NCN) supports fundamental research by funding research projects carried out by individual researches and research teams, both on the domestic and international level, as well as doctoral fellowships and post-doctoral internships. NCN announces calls for proposals four times a year. The NCN grantee must be employed at a Polish host institution.
OPUS is a funding opportunity intended for a wide range of applicants. The research proposal submitted under this scheme may include the purchase or construction of research equipment.
In May 2021, National Science Centre has approved and subsidized three new projects from IPPT PAN:
1. Project: REINFORCEMENT LEARNING FOR SEMI-ACTIVE STRUCTURAL CONTROL AND DECENTRALIZED MITIGATION OF VIBRATIONS: DEVELOPMENT OF NEW CONTROL ALGORITHMS AND ASSESSMENT OF THEIR EFFICIENCY
coordinator: prof. Łukasz Jankowski
Project goal. We intend to develop and verify a fundamentally new and unexplored approach to structural control by:
- Customizing, developing and applying the machine learning techniques of reinforcement learning (RL) for the purpose of structural control.
- Investigating and optimizing their performance and robustness in semi-active global and decentralized mitigation of structural vibrations.
- Verification of the developed methods in numerical and experimental setups.
Motivation. Active structural control can be summarized as working against the structure by applying external control forces. This is potentially dangerous. In contrast, recently surging semi-active control is based on a fundamentally different, nature-inspired paradigm of dynamic self-adaptation: it relies on low-cost adaptation of local structural characteristics. The techniques currently employed to design structural control systems are overwhelmingly classical and analytical. They are time-proven and efficient in applications to active control of linear systems. However, their effective application is much more difficult in case of semi-active control. This is due to the nature of the semi-active actuators, which are energy-efficient and often fail-safe but hard to be effectively processed using classical formulations. There is thus a clear need for new, effective and robust techniques applicable in semi-active control of structures, and this is the main motivation behind this project.
Reinforcement learning (RL), a field of machine learning, is founded on the idea of learning from interaction with the environment. RL approaches achieved recently exceptional successes in a variety of hard problems, ranging from the superhuman level of proficiency in the games of chess and Go, through thermal soaring of gliders and swimming by body undulation, to autonomous car driving. The algorithmic framework of RL provides two features (trial-and-error search and the ability of handling delayed rewards) that make it perfectly suited for tackling hard problems of structural control. Astonishingly, RL seems to be largely ignored in structural control. We are aware of only a handful of attempts which, although pioneering, are limited and involve active control, simple structures or simple RL approaches. This project aims at bridging this gap.
Planned research.Instead of designing control algorithms ourselves, we will design a framework that learns them by itself in repeated trial-and-error interactions with simulated virtual environments. We will start with simple structures and simple control aims to form basic algorithmic building blocks for the subsequent tasks. Then, we will consider decentralized control of modular structures and aim to generalize the control algorithms to be effective in various geometric configurations of modules and to rely on the signals from their local neighborhood only. This should lead to plug-and-play type of controlled modules. Finally, we will promote robustness, that is ensure the control is enough error-tolerant to cope with real-world structures. To achieve these aims, we will adapt, propose and utilize various concepts: ensemble learning, collective control, actor-critic architecture, etc. The performance will be assessed numerically and experimentally, and compared to classical control.
Substantial results expected.Structural vibration is a ubiquitous engineering problem. The project will contribute to the advancement of cost-effective, lightweight and safe structures by development of new RL-based frameworks for robust semi-active structural control. In particular, we expect the project to result in
- RL framework for learning general robust control algorithms applicable at the global and local levels of complex structures, including modular structures.
- A number of specific semi-active control algorithms.
- Specyficzne techniki RL promujące uogólnione, odporne algorytmy sterowania (uczenie i sterowanie zespołowe).
We expect the project to generate original high-quality results publishable in world leading scientific journals in the fields of structural mechanics, smart structures and structural control.
2. Project: NEW GENERATION OF LOVE WAVE BIOSENSORS AND CHEMOSENSORS WITH GIANT SENSITIVITY BASED ON METAMATERIALS
coordinator: prof. Piotr Kiełczyński
In the proposed project, the author seeks a solution for new emerging theoretical and practical problems, resulting from the present status of ultrasonic sensors and recent developments of new revolutionary type of materials, called metamaterials. In the proposed project, the author seeks a solution for new emerging theoretical and practical problems, resulting from the present status of ultrasonic sensors and recent developments of new revolutionary type of materials, called metamaterials. Most modern ultrasonic sensors of physical quantities (e.g., viscosity), biosensors and chemosensors employ mechanical surface waves of the Love type. The maximum mass sensitivity of the existing Love wave sensors (S≈500 m^2⁄kg), approached currently a saturation level, since its magnitude was mainly achieved by increasing the operating frequency of the sensor to extremely high values ~1 GHz. Further augmentation of the frequency is not an option. To resolve this deadlock the author of the project proposes employment of a completely new theoretical and experimental approach by application of recently discovered novel revolutionary materials, called metamaterials, with their extraordinary properties, such as negative mass density and modulus of elasticity. We anticipate that Love wave sensors based on metamaterials should be many times (100-1000) more sensitive at 1 MHz, than the existing Love wave sensors, at operating ~500 MHz. The scientific objective of the proposed project is to develop theoretical foundations and mathematical model for Love surface waves propagating in layered metamaterial waveguides, forming therefore a basis for further development of novel Love wave sensors with extremely high sensitivities. This will constitute a breakthrough that can revolutionize the design and construction of sensors, biosensors and chemosensors working in liquid and gaseous environment. The proposed project represents therefore a pioneering contribution to the basic metamaterial research and the technology of ultrasonic sensors. The theory of Love surface waves propagating in layered metamaterial waveguides (Direct and Inverse Sturm-Liouville Problems) was not yet presented in the literature, therefore the proposed project constitutes a world-wide novelty, from the theoretical, experimental and basic metamaterial research point of view. Metamaterials are characterized with completely unexpected and extraordinary new properties, not existing in conventional materials, such as negative effective mass density with tensorial character, negative effective elastic moduli of elasticity, negative Poisson ratio, invisibility (cloaking), super-focusing, super-resolution, reverse Doppler effect, transmission through ultra-narrow channels, perfect wave absorbers, etc.. Love waves are elastic surface waves propagating in waveguides composed of an elastic surface layer deposited on an elastic substrate. If loaded on top with a viscoelastic liquid or mass density ∆σthe phase velocity and attenuation of the Love wave will change by ∆v and ∆α. The coefficient of sensitivity of the Love wave sensor, with respect to mass loading, is defined as S = ∆v/∆σ/v. The Author of the Project put forward the following research hypotheses: 1. Love wave sensors based on metamaterials can achieve giant sensitivities at reasonable low operating frequencies ~1 MHz, (not at 500 MHz) 2. the power flow P1 in the metamaterial surface layer and that P2 in the metamaterial substrate is anti-parallel 3. the integrity of the Love surface wave is preserved, the wave front is continuous 4. proper selection of metamaterial parameters will lead to cancellation of the overall power flow P = (P1 + P2) →0 The proposed project is very ambitious and highly innovative with an enormous potential for novel challenging applications, such as detection of nanoscopic amount of viruses and bacteria or residual concentrations of warfare agents and drugs . In the first step, we have to develop the theory for Love surface waves propagating in elastic metamaterial waveguides from scratch (basic research). Next, we will have to address many unanswered questions, about possible limits of the phase and energy velocity of the Love wave(can group velocity →0 ?), the directions of the power flow in the metamaterial surface layer and metamaterial substrate, influence of losses on dispersion curves, the overall power flow P, etc.
3. Project: CELL-TO-CELL HETEROGENEITY OF CROSS-WIRED CYTOKINE SIGNALING: FILLING THE GAPS BETWEEN MOLECULAR ORIGINS AND TRANSLATIONAL IMPLICATIONS
coordinator: prof. Michał Komorowski
In multicellular organisms like the human body, trillions of cells, of multiple different cell types communicate with each other by releasing a thousand types of molecules such as hormones, growth factors, cytokines, or chemokines to coordinate an organism in all aspects of its function. The functioning of biochemical signaling, however, does simply not comply with the principles of communication engineering. For instance, an engineer designing a communication system would use few distinct signaling components while ensuring that the output of each component is highly reproducible. Natural evolution came up with a different solution: cells have many interconnected, cross-wired pathways that produce highly variable output signals. How can cells function reliably with highly variable signaling outputs? What is the explanation for cross-wired architecture? What are the implications of variable cross-wired signaling for health and disease? These questions reflect tangible gaps in our understanding of how cellular signaling functions.
In the prevalent view, the highly variable signaling outputs of single-cells result from stochasticity (molecular noise)—and not deterministic factors—and therefore diminish signaling fidelity. In our recent work, using bi-nuclear syncytia, we showed that less than 10% of the cell-to-cell heterogeneity of primary signaling outcomes can be attributed to noise along the signaling pathway inside the cell. The remaining 90% of cell-to-cell heterogeneity arises from phenotypic variability. Further, in our earlier theoretical work, we suggested that expansion of signaling components, be it receptors or signaling effectors, via duplication and enlistment of promiscuously acting cues is virtually the only accessible evolutionary strategy to achieve overall high-signaling capacity despite overlapping specificities and molecular noise. This mode of expansion explains the highly cross-wired architecture of signaling pathways. In addition, we have recently used probabilistic modeling and information-theory, to introduce fractional response analysis (FRA), which enables a systematic investigation of cell-to-cell heterogeneity structure for multivariate and high-throughput data in a wide range of situations, in which response analysis in single-cells is of relevance. The main goal of this proposal is to follow the research avenues opened by our recent work, in order to address the specific long-lasting problems in signaling:
- Aim 1: Characterise determinants of cell-to-cell heterogeneity of cytokine responses;
- Aim 2:Uncover evolutionary forces that might have shaped cross-wired signaling;
- Aim 3: Examine the impact of aging on cell-to-cell heterogeneity cytokine response;
- Aim 4:Examine the impact of cell-to-cell heterogeneity on the efficacy of single-cell inhibition in EGF signaling dependent cancer cells.
The exact implications of cell-to-cell variability and cross-wired architecture of signaling pathways for understanding of cells’ functioning as well as for etiology and treatment of human disease are only beginning to emerge. For instance, recent evidence, revealed that cell-to-cell variability of gene expression increases during aging. Besides, in pharmacology, the cell-to-cell variability acts as a catalyst for a paradigm shift in preclinical studies, by increasingly incorporating single-cell measurements. In cancer, for instance, fractional killing or incomplete growth inhibition of clonal tumor cells is a significant problem, which cannot be inspected with population measurements. The proposed research agenda is aimed to fill certain gaps in our understanding how cell-to-cell heterogeneity and cross-wired architecture should be accounted for when aiming to understand functioning of cellular signaling systems. In particular, we expect to explain what factors determine cytokine responses of individual cells, as well as provide insights into what forces might have cross-wired architecture of signaling. In addition, our results are intended to demonstrate how aging changes cell-to-cell heterogeneity structure of cytokine responses in human immune cells so that in the future, cell-to-cell heterogeneity could be developed into a quantitative trait comparable across individuals, and a biomarker for aging, or immune vulnerability. Furthermore, our experiments may provide valuable insight that can further inform the preclinical development of effective therapeutic interventions in signaling.
