AI drives motorcycle innovation

About the customer

Hero MotoCorp Ltd. (Hero), a multinational motorcycle and scooter manufacturer headquartered in Delhi, India, serves millions of customers across more than 45 countries. As one of the world’s largest motorbike manufacturers, Hero holds nearly 31 percent of the Indian two-wheeler market and has established partnerships with global brands like Zero Motorcycles and the iconic Harley-Davidson. Hero strives to optimize development by using best-in-class technology and the latest simulation methods – including artificial intelligence (AI) and machine learning (ML).

Their challenge

Simulation has been a central part of Hero’s product development for years; traditional finite element analysis (FEA) has enabled Hero’s development teams to perform component design evaluations and provide detailed design insights. However, Hero sought to overcome development bottlenecks and drive even quicker decision-making. This led them to move beyond traditional FEA and implement AI and ML. To support their computer-aided engineering (CAE) engineers in adopting AI and ML in their daily workflows, Hero needed efficient, powerful, user-friendly tools. Hero’s first AI-driven project focused on designing a new motorcycle handlebar. The handlebar is a vital motorcycle component; it influences ergonomics, defines the rider’s position and impacts the motorcycle’s steering, balance, comfort and character. Because of its prominent role, teams typically invest a lot of time designing and refining the handlebar. To meet their goals and shorten the design process, Hero turned to Siemens’ AI-powered technology.

Handlebar stress results from traditional FEA method.

Our solution

Hero chose Simcenter™ PhysicsAI™ software, a powerful geometric deep learning technology that leverages historical data to train AI models and generate physics predictions in a fraction of the time that traditional FEA methods require. The Simcenter PhysicsAI workflows, conveniently embedded in Simcenter, Siemens’ simulation and design platform, enabled Hero users of all skill levels to seamlessly weave this solution into their existing processes.

Simcenter is part of the siemens Xcelerator business platform of software, hardware and services.

To leverage Simcenter PhysicsAI, Hero first split their existing data into training and testing sets: training data to train their ML model on past simulations, and test data to evaluate and quantify the AI model’s predictive accuracy. With a wide range of bikes – sport bikes, adventure bikes, commuter bikes, cruisers, etc. – Hero’s team used a variety of handlebar datasets to ensure the AI model produced accurate results.

Data segregation and model training are critical steps in the AI-powered engineering process. Hero opted for a typical 80/20 data split, meaning 80 percent of the data was specified for model training and 20 percent was for evaluating predictions. The project team selected 30 handlebar variations for training from 24 datasets. After training, they tested the AI model with the remaining six handlebar datasets and evaluated the accuracy of the results by comparing traditional FEA results to the AI-generated predictions. The comparison revealed only a 3 percent variation, proving that Simcenter PhysicsAI delivered accurate predictions in a fraction of the time compared to traditional FEA.

In a separate simulation, the team imported a handlebar dataset that wasn’t part of the training or testing sets to predict stresses on new handlebar designs. Using their newly trained Simcenter PhysicsAI model, the team could predict stress results with a single click. They compared the Simcenter PhysicsAI predicted stress data to the values in a stress plot based on traditional FEA results, which once again revealed a less than 3 percent variation. The close match gave engineers confidence in the AI-generated results for a new design concept. The approach also saved time.

“While our traditional FEA simulation method took a full hour, Simcenter PhysicsAI generated similar predictions – within a 3 percent result variation – in just three minutes,” says Jeevesh Prasoon, senior engineer at Hero MotoCorp.

Stress results from Simcenter PhysicsAI compared to traditional FEA show a 3% variation.

Results

Using Simcenter PhysicsAI provided enormous benefits for Hero. Most notably, the time savings – from an hour to mere minutes – were substantial. Additionally, the Simcenter PhysicsAI simulation was accurate, with its results showing only a 3 percent variance compared to traditional FEA analysis results. Applying AI-powered engineering processes helped Hero reduce product development time, enrich its team’s skill set and accelerate design exploration.

“Simcenter PhysicsAI significantly reduced our development time,” says Prasoon. “With it, we can generate conclusive results at lightning speed and accelerate our decision-making.”

Based on this success, Hero plans to scale its use of AI technology, using it to design other complex components for even greater efficiency, accuracy to enhance decision-making.