With regards to making use of machine studying to bodily system modeling, it’s an increasing number of widespread to see practitioners shifting away from a pure data-driven technique, and beginning to embrace a hybrid mindset, the place wealthy prior bodily information (e.g., governing differential equations) is used along with the info to enhance the mannequin coaching.
Below this background, physics-informed neural networks (PINNs) have emerged as a flexible idea and led to many success tales in successfully fixing real-world challenges.
As a practitioner who is keen to undertake PINNs, I’m eager on studying each the newest developments in coaching algorithms, in addition to the novel use instances of PINNs for real-world functions. Nevertheless, a ache level I usually see is that, though there are ample analysis papers/blogs summarizing efficient PINN algorithms, overviews of novel use instances of PINNs can hardly ever be discovered. One apparent purpose is that, not like the coaching algorithms that are domain-agnostic, reviews of PINN use instances are scattered in numerous engineering domains and never readily accessible for a practitioner who’s normally an skilled in a single particular area. As a consequence, I usually discovered myself reinventing the wheel as my methods of utilizing PINNs have already been effectively addressed by practitioners in one other subject.
It’s precisely my journey and experiences which have sparked the thought of penning this weblog: right here, I attempt to interrupt the knowledge barrier throughout totally different engineering domains and distill the recurring useful utilization patterns of PINNs. I hope that this assessment will inform practitioners from totally different domains about what’s attainable with PINNs and encourage new concepts for interdisciplinary innovation.
Towards that finish, I’ve extensively reviewed PINN analysis papers up to now three years and got here up with the next 5 most important utilization classes:
- Predictive modeling and simulations
- Optimization
- Knowledge-driven insights
- Knowledge-driven enhancement
- Monitoring, diagnostic, and well being evaluation