MBD Simulation AI Platform for Automotive

A platform that runs MATLAB / Simulink control models and simulations in parallel on Azure / Kubernetes, supporting verification, parameter exploration, and code generation from the specification stage.

Challenge

In conventional development, evaluation required building physical prototypes, so every defect caused large rework. Implementing software meant reading through massive specifications, which invited misreads and bugs, and some evaluation cases could not be tested physically on real hardware.

Solution

Built a platform that runs MATLAB / Simulink control models and simulations at scale on Azure, based on Model-Based Development (MBD) and CAE. Used Kubernetes (AKS) to run analysis jobs in parallel, connecting control-model verification, parameter exploration, and automated result aggregation from the specification stage. Models run as executable specifications, enabling automatic code generation from models and simulation of evaluation scenarios that are hard to reproduce on real hardware.

Result

Teams can verify control-model behavior at the specification stage, finding and fixing most defects during modeling. Rework after software implementation is minimized, and a wider range of evaluation scenarios improves reliability.

Team

1 member, PoC phase

Simulation infrastructure and MBD workflow design

Role

Handled simulation infrastructure, MBD/CAE workflow design, and MATLAB/Simulink integration.

Connected control-model execution, parallel analysis jobs, and result aggregation into a single workflow on Azure.

Tech Stack

ModelingMATLAB / Simulink(MBD・CAE)
ComputeAzure (AKS) / Kubernetes
Orchestration並列解析ジョブ / パラメータ探索
BackendPython / REST API
InfrastructureMicrosoft Azure

Key Features

01

Run MATLAB / Simulink control models and simulations in parallel as analysis jobs on Kubernetes (AKS)

02

Verify model behavior at the specification stage, before writing implementation code

03

Parameter exploration and automatic aggregation of simulation results

04

Automatic code generation from verified models

05

Simulate evaluation scenarios that are hard or impossible to reproduce on real hardware

Technical Highlights

Catch Defects During Modeling

Because control models are built when the specification is fixed, behavior can be verified before coding, so most defects are found and fixed at the modeling stage and post-implementation rework is minimized.

Models as Executable Specifications

Running models as living specifications removes spec misreads, smooths implementation, and allows programs to be generated directly from the models.

Broader, More Reliable Evaluation

Simulation covers evaluation cases that cannot be tested physically on real hardware, widening coverage and improving reliability.