Ashok Leyland - Product Intern | Sanjana Kanchibotla

The Problem

Testing engines is expensive. Every physical test requires setting up the engine on a test bed, running it for hours, burning fuel, and having engineers monitor instrumentation. The R&D team needed a way to predict how engines would perform under different conditions without running all those tests.

What I Did

Built a simulation model from scratch

I used GT Suite (GT Power) to build a thermodynamic model of the H64v engine - a 6-cylinder, 4-valve commercial diesel engine in 250 HP and 280 HP configurations. This meant modeling the entire system: turbochargers, intercoolers, EGR systems, fuel injection, exhaust manifolds, everything.

Calibrated it against real test data

I validated the model in two ways. For steady-state conditions (constant speed and load), I tested full throttle across 800-2600 RPM and part throttle at 10 different load points. For transient conditions (changing speeds and loads), I validated against real duty cycles like T90, Sweep, Lug, and WHTC. The trick was matching the simulation to actual test bed measurements at critical points.

Ran parametric studies

Once the model was accurate, I tested variables like compression ratios, valve lifts, and injection timing to see their impact on power and efficiency. This helped the team make design decisions for the 280 HP variant without physically building and testing every configuration.

Results

Steady-state performance:

Brake torque error within ±3%
Boost pressure error within ±0.2 bar
Fuel consumption error within ±5%
Overall model accuracy >97% across all parameters

Transient performance:

T90 timing error within 0.3 seconds
Torque and fuel consumption error within ±4%

The model was accurate enough that engineers could confidently make design decisions without physical testing. I received a Certificate of Appreciation from the Senior Vice President for the work quality.

What I Learned

95-97% accuracy sounds imperfect, but it's exactly what was needed. Perfect accuracy would've required more computational power and time than it was worth. The model just had to be good enough for engineers to trust the predictions over running expensive physical tests. Balancing accuracy with practicality is more important than chasing perfection.