Conventional Artificial Intelligence (AI) and Machine Learning (ML) models usually used in the IT industry relies on the availability of large amount of data collected over a period of time. However, in engineering, in general, one cannot use a large number of sensors on the system to obtain large amount of data in real time. Digital Twins in engineering, therefore, relies on development of sparse data analytics integrated with the physics-based responses of the system to operating loads in real time. Generally, these are in-situ response data based on Newtonian laws obtained either experimentally, analytically, numerically or a combination of them. Coupling sensor data with the physical behavior of a system under operating loads will provide a better insight into the systems behavior for its effective control and maintenance. In addition, uncertainties always exist when physical systems are being used in the general environment. Probability integrated into the physics-based models using ML and deep learning methodologies help in further understanding of the system behavior to unknown inputs.

At QuantumPoint, we have expertise in the development of actual digital twin hardware integrating AI and ML along with the real time physics-based models. As an example, if one considers the rotor dynamic behavior of a typical motor coupled to a shaft or a pump or a gear box, the digital twin developed actually looks at the acceleration response, current response, temperature response etc. We analyze more than one physical response for better understanding and prediction of the system behavior. In this example, we look at the particular frequency response behavior of a typical acceleration response and build in the intelligence. We follow Newtonian physic approach: for example, we would analyze if at a particular frequency the acceleration response is showing a high amplitude. Such a understanding would help identify to a certain degree of probability that this event corresponds to a particular defect. In case of rotating machines such defects could be related to parallel misalignment, angular misalignment, bearing defects etc. This is possible if physics is integrated to analytics. Our engineers know what has failed, how it has failed and why it has failed. Also, our engineers know if the system can operate without a catastrophic failure for some more time for example. This will help predict the remaining useful life including both probability and reliability techniques in the ML algorithms. We ensure to achieve looking at the physics of the response in real time. Our vast experience in this area suggests, in engineering most data are obtained from outside by positioning sensors. However, most failures happen inside. Unless, the physics of the behavior is integrated into analytics, it is difficult to predict the inside failure using outside data.

QuantumPoint’ expertise in physics-based analytics also help to develop tools and technologies for both diagnosis and prognosis using minimum data and integrating the physical behavior. We also help to validate numerical simulations with experimental data to enhance customer’s digital twin solutions to become more robust in terms of its capability in diagnosis and prognosis. We also include probability-based uncertainty quantification tools on the top of this, make better tuning your ML to be used as a robust tool in real-time.