Digital Twin Systems

Maintainer: Lee Bissessar.
GitHub Repo: https://github.com/digitaltwin-ai-org
A digital twin system is a cutting-edge technology that leverages the power of artificial intelligence (AI) to create virtual replicas of physical objects, processes, or systems. It involves the integration of various data sources, such as sensors, Internet of Things (IoT) devices, and historical information, to create a comprehensive digital representation of the real-world counterpart. This digital twin serves as a dynamic and interactive model, providing real-time insights and simulations for analysis, monitoring, and control purposes. By bridging the physical and digital realms, digital twin systems enable researchers and engineers to gain a deeper understanding of complex systems, optimize performance, predict behavior, and even facilitate remote teaching and control through AI algorithms. The combination of digital twin systems and AI technologies opens up new possibilities for enhancing efficiency, improving decision-making processes, and driving innovation in a wide range of domains, from manufacturing and healthcare to transportation and smart cities.
In our preliminary research, we have successfully developed digital twins of various complex systems, including a quadruped robot, a self-landing rocket, a drone, an underwater autonomous vehicle, and a Mars rover. These digital twins have provided us with valuable insights into the behavior, performance, and control of these systems, enabling us to optimize their operations and enhance their capabilities. Building upon this success, we now intend to extend our research into the realm of process and power systems. By developing digital twins for industrial processes, power plants, and energy grids, we aim to leverage the power of AI to analyze and optimize their efficiency, predict failures, and implement advanced control strategies. This expansion of our research will enable us to unlock new possibilities for enhancing sustainability, reliability, and productivity in critical domains, ultimately contributing to advancements in energy management and process optimization.

Algorithms: Internal Cogitation

Prompt engineering refers to the process of carefully crafting and refining prompts to elicit specific responses or behaviors from language models like ChatGPT. It involves formulating clear instructions, providing context, and using techniques such as question-answering formats, completion prompts, or conversation simulations to guide the model's output towards desired outcomes. Methods such as few-shot, chain-of-thought and tree-of-thought prompting, can be thought of as external mechanisms to guide AIs to cogitate on problems to achieve the desired answer. In this research, we aim to develop these methods internally to the network, resulting in faster more accurate responses.

---

Algorithms: Coherence

In this research, we demonstrate the alignment between human and machine understanding while proposing quantitative metrics and highlighting the associated benefits of this alignment.

---

Applications: AI enabled Industry

The emergence of Industry 4.0 has ushered in a new era characterized by the generation of massive volumes of data, paving the way for seamless integration between AI systems and SCADA systems. This integration empowers process systems to perform efficient analysis of alarms, effectively handle abnormal situations, accurately assess asset lifespan, and proactively address potential cybersecurity threats. Presently, we are actively engaged in collaborations with both local and international oil & gas companies as well as service providers to implement these innovative solutions, ensuring their practical application and maximizing their impact across the industry.

---

Applications: AI in Education

This research focuses on two pathways. Firstly, we investigate the necessary conditions to ensure desired learning outcomes for students utilizing generative learning systems, while also addressing potential challenges that arise from their misuse. Secondly, we develop AI systems integrated with pedagogical concepts, aiming to facilitate student learning through authentic assessments and problem-solving activities that are centered around the students themselves. By exploring these two distinct paths, we aim to enhance the understanding of effective educational practices in the context of technology-driven learning environments and promote student engagement and success.

---

Intersampled Iterative Learning Control

Iterative learning control is a control strategy that aims to improve the performance of systems through repeated iterations of control actions. It leverages the knowledge gained from past iterations to refine and enhance future control actions, leading to reduced errors and improved tracking accuracy over time. This approach is particularly effective for repetitive tasks or processes where the system learns from its own history to achieve desired outcomes. In 2014 we developed a family of control laws that collapsed the iteration domain into the time doamain. In this case an internal model (albeit imperfect) of the system is stored in the controllers memory and iterated on several times in a virtual "practice makes perfect" way before output. This guarantees monotonic zero error convergence and minimizes the long term expected errors.

---

Reinforcement Learning

This research is focused on a class of reinforcement learning strategies with attention mechanisms, restricted exploration and better optimization techniques to produce more general Neurodynamic strategies.

---

Attack Resiliency

Attack resiliency, in the context of cyber-physical systems and control systems, refers to the ability of a system to withstand and mitigate the impact of malicious attacks or unauthorized intrusions. It involves implementing robust security measures, proactive monitoring, and adaptive response mechanisms to detect, prevent, and recover from potential cyber-attacks. Attack-resilient systems are designed to maintain their functionality, protect critical infrastructure, and ensure the safe operation of control systems even in the face of persistent threats, minimizing the potential damage caused by malicious activities.
We have developed intelligent controllers based on fuzzy and RL algorithms for rejecting attacks on intelligent vehicle platoons.

---

Intelligent Control of Decentralized Robot Systems

Intelligent control of decentralized robots refers to the implementation of advanced algorithms and decision-making techniques that enable a group or swarm of individual robots to work collaboratively and autonomously. Instead of relying on a centralized control system, each robot in the decentralized system possesses intelligence and autonomy to make independent decisions based on local information and communication with neighboring robots. Through coordination and cooperation, these intelligent decentralized robots can perform complex tasks, adapt to dynamic environments, optimize resource allocation, and exhibit emergent collective behavior. This approach allows for flexible, scalable, and fault-tolerant robotic systems that can efficiently tackle challenges in various domains, such as swarm robotics, multi-robot systems, and distributed sensing applications.
This research focuses on pose control of an object manipulated by a deformable sheet supported by multiple robots.

---

Deep Learning for Electric System Asset Management

---

COVID-Tech

---

Smart Agriculture Systems

---