We are proud to announce our participation in the PIAI4Pumps project, recently developed in collaboration with the MADE competence center, which further strengthens our commitment to digital innovation. More specifically, PIAI4Pumps represents a concrete step forward for Trillium toward faster, more structured, and data-driven centrifugal pump design, thanks to the integration of advanced numerical simulation and physics-informed artificial intelligence.

Below you can find the full project official details.

TRILLIUM PUMPS ITALY S.P.A. – PIAI4PUMPS

Trillium Pumps Italy S.p.A., a joint-stock company with share capital of €21,000,000.00 headquartered at Via Gaetana Agnesi 1, 20834 Nova Milanese (MB), Italy.

Partner in the MADE* project that worked on PIAI4Pumps – Physics-Informed Artificial Intelligence for Rapid Simulations of Fluid Flows in Centrifugal Pumps.

Trillium Pumps Italy S.p.A. operates in the field of centrifugal pumps and fluid handling solutions, serving industrial environments where reliability, hydraulic performance, and efficiency are key competitive factors. The company develops and applies design and numerical simulation expertise to the operation of centrifugal pumps, with the aim of improving performance, shortening development times, and increasing product innovation capacity.

THE CHALLENGE

The design of centrifugal pumps requires complex computational fluid dynamics simulations based on models capable of describing turbulent phenomena and providing performance indicators useful for product development. These activities involve long computation times and make it costly to explore multiple design scenarios, which can limit optimization opportunities and slow down the design process. Existing AI strategies also do not effectively account for highly turbulent flows such as those typically encountered in centrifugal pump development. For this reason, Trillium identified the need for a methodology capable of making simulations much faster without losing physical consistency in the results.

THE SOLUTION

The PIAI4Pumps project is aimed at developing a computational procedure for solving fluid dynamics problems in centrifugal pumps through innovative Physics-Informed Machine Learning techniques. The approach uses data obtained from numerical simulations to train neural network-based architectures capable of learning and reproducing, in a physically consistent way, the input-output relationships relevant to pump design. The solution combines dimensionality reduction methods to compress simulation data, neural models to learn velocity and pressure fields, and multi-fidelity methods to integrate experimental data and numerical outputs, where available, in order to further improve accuracy. The project is structured in three phases: analysis of the AS-IS process and scientific literature, design and configuration of the PIML model, and training and validation of the model with the release of an application prototype.

TECHNOLOGIES

The project’s core technologies include Physics-Informed Machine Learning for fluid dynamics problems in turbulent environments, dimensionality reduction techniques such as Proper Orthogonal Decomposition and possible convolutional autoencoders, as well as surrogate models based on neural networks or Gaussian processes. The project also uses training datasets built from numerical simulation campaigns and multi-fidelity approaches to combine experimental and numerical data within machine learning pipelines.

IMPACT

The project aims to make numerical simulations up to 1,000 times faster, while accepting a less exact approximation of the result but preserving practical usefulness for design and optimization activities. This advantage makes it possible to address optimization problems that were previously unsolvable and to drastically reduce the time required for analyses that currently take several days. The initiative also highlights a positive environmental sustainability impact, since reduced computation leads to lower energy consumption. Looking ahead, structuring design knowledge into reusable datasets and models strengthens the company’s innovation capacity and lays the foundations for further future developments.

BENEFITS FOR THE COMPANY

For Trillium, the project offers direct benefits in terms of faster design assessments, the ability to analyze a greater number of pump configurations, and more effective support for development decisions. The creation of a structured dataset for use in machine learning pipelines, together with the availability of a software prototype and a surrogate model application, generates technical assets that can be immediately leveraged within the company. The project also supports better capitalization of designers’ know-how, which will be useful for future improvements to the AI model and for new industrial applications.

KPIs

• Numerical simulation speed-up of up to 1,000 times compared with the traditional approach.
• Creation of a structured dataset for use in machine learning pipelines.
• Development and validation of a software prototype and a surrogate model application.

 

(*)MADE s.c.a.r.l. (MADE) is one of the competence centers which, following the Director’s Decree of the Italian Ministry of Economic Development dated 29 January 2018, was recognized as one of the centers of excellence for the development of orientation and training activities, as well as innovation, industrial research, and experimental development projects on Industry 4.0 topics.