Danfoss - Enhancing Check Valve Design with Deep Learning
Danfoss is currently collaborating with Neural Concept to evaluate the integration of Neural Concept Shape (NCS) deep learning technology within their product design workflows. The technology has been initially assessed in the context of the design of a new check valve.
NCS models were used to predict the performance of new valve designs by evaluating the mass flow rate, the pressure drop, and the lifting force on the piston, as well as the steady 3D flow field inside the check valve, which the technology provides within seconds.
About 100 designs and their corresponding CFD simulations were used for the model training. The relative error between predicted quantities on unseen geometries and those obtained through CFD simulations was about 1%.
Bottom: static pressure on central cross-section of the fluid volume, on a test sample
NCS’ Neural Design module was used for creating new designs, in a smart, innovative, non-parametric way.
A combination of the tools mentioned above was used to perform a multi-objective optimization. In total, 750 designs have been assessed in 15 optimization iterations consisting of 50 new designs each, with an overall turnaround time of 10 hours on a GPU.
The optimization led to the exploration of a very wide design space and an approximately 10% improvement of the mass flow rate at check valve's fully open condition.
About Danfoss
Danfoss engineers solutions that increase machine productivity, reduce emissions, lower energy consumption, and enable electrification. Our solutions are used in such areas as refrigeration, air conditioning, heating, power conversion, motor control, industrial machinery, automotive, marine, and off- and on-highway equipment. We also provide solutions for renewable energy, such as solar and wind power, as well as district-energy infrastructure for cities. Our innovative engineering dates back to 1933. Danfoss is family-owned, employing more than 42,000 people, serving customers in more than 100 countries through a global footprint of 95factories. www.danfoss.com
About Neural Concept
Neural Concept develops DeepLearning tools and algorithms for enhanced engineering. By speeding up R&D cycles, augmenting product performance, and solving their next engineering challenges, we revolutionize how our customers design their products. The company was founded in 2018 in a top-notch AI research lab at EPFL in Switzerland. We are a team of 35+ members fully committed to our vision, which is to revolutionize the future of industrial engineering with intelligence.
