Periodic Reporting for period 1 - RealFlex (Real-time simulator-driver design and manufacturing based on flexible systems)
Período documentado: 2019-10-01 hasta 2021-09-30
Virtual prototyping is one of the key elements of modern design, training, and maintenance of machines. However, virtual prototyping requires reliable, highly-accurate, simulation solutions. Modern simulation software, like Mevea simulator, includes multiple phenomena crucial in machine design and testing. Physics-based rendering of such phenomena is possible, however, it put a lot of burden on computer. It is well-known that large, high-accuracy models often require hours or days to compute. However, when virtual machine is operated by a human driver, simulation must response to operators’ actions in real time (therefore they are called real-time simulation). Realistic, real-time simulation solutions for machine simulators is one of the areas where Mevea truly stands out.
The main objective of the project is to enable reliable, effective, efficient, and general solution for modeling deformable bodies in real-time simulations done by Mevea software. This is done by developing and implementing state-of-the-art methods and research on novel approaches.
Effective and highly accurate real-time simulations have many important uses. In virtual training, operator or driver can master its skills effectively. In modern designs, better products can be delivered even faster. Autonomous vehicles require good models in their control algorithms for operation. Even after production, in the real world, combined technologies like digital twins, smart sensors, edge computing, and internet of things, can ensure safe and smooth operation of the machine. All this translates into advantages important for the society: limited resource consumption, time and energy savings, cost efficiency, and improved health safety.
The main objective of the project is to enable reliable, effective, efficient, and general solution for modeling deformable bodies in real-time simulations done by Mevea software. This is done by developing and implementing state-of-the-art methods and research on novel approaches.
Effective and highly accurate real-time simulations have many important uses. In virtual training, operator or driver can master its skills effectively. In modern designs, better products can be delivered even faster. Autonomous vehicles require good models in their control algorithms for operation. Even after production, in the real world, combined technologies like digital twins, smart sensors, edge computing, and internet of things, can ensure safe and smooth operation of the machine. All this translates into advantages important for the society: limited resource consumption, time and energy savings, cost efficiency, and improved health safety.
The project starts with research on deformable body modeling techniques that can be used in real-time software. As the main objective of the project is to introduce general approach to deformable body modeling, a well-developed floating frame of reference was selected upfront. Next step was the selection of the deformable body representation. Research shows that there exists open format for description of flexible bodies using FFRF, support many reduction methods and is supported by commercial software’s. This format is called standard input description (SID) file and was chosen for implementation in Mevea.
The practical aspects of including flexibility into the Mevea software includes three areas of development: export of the model from finite element package, middleware that generate SID file, and Mevea solver itself where SID file is read and used in simulation. ANSYS finite elment package was chosen as basis for representation of the flexible body required for FFRF. A dedicated workflow was created to export model data. For the next stage a middleware was created that transform and write deformable body data as SID file. A novel matrix notation was adopted to simplify and improve inertia derivations. Matrix notation enables more exact representation of inertia than traditional index notation. The last element was to implement flexible body support in Mevea. This boils down to importing SID file, configuring flexible body, and computing body forces and points locations. After that stage Mevea supports general flexible bodies modeled using floating frame of reference formulation.
Project includes a secondment at LUT University. During this time two main research activities were carried out. First one was research on novel, special purpose flexibility solution to model elastic ropes. Rope modeling is challenging and standard FFRF solution does not work here. Research results with improved damping model in so-called Absolute Lagrangian-Eulerian Modal formulation. Second research area was in machine learning applications to optimize and control machines in simulation environment. Within this task, a realistic model of the excavator prepared in Mevea software was trained using state-of-the-art reinforcement learning algorithm. As the result, excavator was trained to perform typical earth-moving operations.
The main result of the project is working implementation of the deformable body in the Mevea solver. New virtual machines can achieve higher accuracy, thus improving operator experience. Implementation was tested with standard flexibility benchmarks and on crane example. Another important achievement is working interface for the ANSYS finite element software.
Other exploitations are created research software. First one is written in Python for internal Mevea usage and allow to examine flexible body behavior outside main solver. Second tool is written in Matlab language and allow to analyze rope-pulley systems. Third tool uses Mevea’s Python interface and allow for machine learning training of the artificial agents on the virtual models developed in Mevea.
Dissemination was affected by the coronavirus pandemic. For example, no conference and no international visits were performed. As a result, dissemination was done locally at Mevea and LUT University. A number of internal reports, discussions, results consultancies, and meetings were created and conducted. At LUT two student course was provided about programming, numerical methods, and dynamic simulation of mechanical systems. Knowledge exchange between Mevea company and LUT University allow to perform a challenging study on reinforcement learning control of complex ground-moving machinery.
The practical aspects of including flexibility into the Mevea software includes three areas of development: export of the model from finite element package, middleware that generate SID file, and Mevea solver itself where SID file is read and used in simulation. ANSYS finite elment package was chosen as basis for representation of the flexible body required for FFRF. A dedicated workflow was created to export model data. For the next stage a middleware was created that transform and write deformable body data as SID file. A novel matrix notation was adopted to simplify and improve inertia derivations. Matrix notation enables more exact representation of inertia than traditional index notation. The last element was to implement flexible body support in Mevea. This boils down to importing SID file, configuring flexible body, and computing body forces and points locations. After that stage Mevea supports general flexible bodies modeled using floating frame of reference formulation.
Project includes a secondment at LUT University. During this time two main research activities were carried out. First one was research on novel, special purpose flexibility solution to model elastic ropes. Rope modeling is challenging and standard FFRF solution does not work here. Research results with improved damping model in so-called Absolute Lagrangian-Eulerian Modal formulation. Second research area was in machine learning applications to optimize and control machines in simulation environment. Within this task, a realistic model of the excavator prepared in Mevea software was trained using state-of-the-art reinforcement learning algorithm. As the result, excavator was trained to perform typical earth-moving operations.
The main result of the project is working implementation of the deformable body in the Mevea solver. New virtual machines can achieve higher accuracy, thus improving operator experience. Implementation was tested with standard flexibility benchmarks and on crane example. Another important achievement is working interface for the ANSYS finite element software.
Other exploitations are created research software. First one is written in Python for internal Mevea usage and allow to examine flexible body behavior outside main solver. Second tool is written in Matlab language and allow to analyze rope-pulley systems. Third tool uses Mevea’s Python interface and allow for machine learning training of the artificial agents on the virtual models developed in Mevea.
Dissemination was affected by the coronavirus pandemic. For example, no conference and no international visits were performed. As a result, dissemination was done locally at Mevea and LUT University. A number of internal reports, discussions, results consultancies, and meetings were created and conducted. At LUT two student course was provided about programming, numerical methods, and dynamic simulation of mechanical systems. Knowledge exchange between Mevea company and LUT University allow to perform a challenging study on reinforcement learning control of complex ground-moving machinery.
Progress in modeling of the flexible bodies beyond state-of-the-art was achieved in novel approach of defining inertia terms of the deformable bodies in the FFRF where directly a matrix calculus were applied. This allow to direct utilization of the distributed model mass. In addition, procedure efficiency was improved. As the large systems often occurs, this may have a significant effect on usability of the proposed solution. Second progress came in the form of robust damping modeling for the rope. This as well has important practical implications as no real-world rope model can operate without damping. Moreover, the additional study of the reinforcement learning control went beyond the applications found in the current literature, as the employed model of machine is complex and realistic what is rare in most applications.
The direct effect of the project is the support of the deformable components in Mevea simulations. This affects all Mevea’s customers, places mostly in Europe area, by providing reliable simulations and training simulators. Mevea’s customers include wide variety of businesses, including mining, earth-moving machinery designers, crane designers, and wood industry. This allow to design and manufacture safer and more reliable machines and allow faster and simpler introduction of skilled worker to the marked. Published papers, amplify this effect to more businesses and markets.
From wider perspective, developed solutions, may affect many areas and technologies: machine design, rapid training, model-based control, autonomous vehicles, edge computing, smart sensors, digital twins, and others. Many of those are still in development or early implementation phase and aims on revolutionizing the whole life cycle of machines and vehicles: design, manufacturing, operation, maintenance, and decommission. For sociate it means safer, more reliable, energy efficient, environmentally friendly, and maintainable products.
The direct effect of the project is the support of the deformable components in Mevea simulations. This affects all Mevea’s customers, places mostly in Europe area, by providing reliable simulations and training simulators. Mevea’s customers include wide variety of businesses, including mining, earth-moving machinery designers, crane designers, and wood industry. This allow to design and manufacture safer and more reliable machines and allow faster and simpler introduction of skilled worker to the marked. Published papers, amplify this effect to more businesses and markets.
From wider perspective, developed solutions, may affect many areas and technologies: machine design, rapid training, model-based control, autonomous vehicles, edge computing, smart sensors, digital twins, and others. Many of those are still in development or early implementation phase and aims on revolutionizing the whole life cycle of machines and vehicles: design, manufacturing, operation, maintenance, and decommission. For sociate it means safer, more reliable, energy efficient, environmentally friendly, and maintainable products.