JTEKT’s Software Factory Cuts Development Time from Days to Minutes
Bringing Continuous Integration into Safety-Critical Vehicle Systems
The automotive industry is under pressure. Manufacturers are accelerating development timelines as the complexity of vehicle software continues to skyrocket. Yet steering systems, where software failures mean potential catastrophe, still require rigorous safety standards. Traditional development methods—with their sequential workflows and manual testing processes—can no longer keep pace with market timeline demands. JTEKT, a global supplier of power steering technology, had to adapt the way it designed and tested the systems. JTEKT Chief of Automotive Business Unit and Executive Officer, CTO, Mr. Sakata, led an initiative to standardize and unify software engineer development processes, methods, and tools.
JTEKT engineers needed a new approach: a software factory. A software factory is a structured methodology for software development that treats software creation like a manufacturing process. It uses standardized tools, templates, and automation to produce software efficiently and consistently.
JTEKT’s transformation from manual processes to an automated model-based software factory enabled the company to dramatically accelerate development cycles while maintaining the ISO® 26262 compliance essential for safety-critical systems. Teams are more efficient, and processes that once took days now complete in under an hour. The approach is proving its worth beyond steering systems.
Steering system. (Image credit: JTEKT)
The Breaking Point
When Nicolas Bertotto, a section manager in the system and software department, joined JTEKT in 2016, the development process reflected the industry norms at that time. The software architecture consisted of large, monolithic components, making every change laborious. Bug fixes could consume days of engineering time.
“Sometimes it could take me a day or two to fix a simple bug,” says Bertotto. “And when I was doing this, progress on my other tasks halted.”
Luc Malrait, a senior software engineer at JTEKT, witnessed how manual processes consumed valuable engineering time. Beyond designing models, engineers had to manually verify compliance with modeling rules, confirm code was bug-free, and ensure all functions had proper test coverage. These tasks created a web of inefficiencies.
The challenge with artifact management: an artifact for one component and artifacts for one project.
Projects contained scattered automation scripts, often with identical names performing different functions. Documentation and test vectors were sometimes out of sync with actual models. Configuration management was so complex that onboarding a new engineer to a project could take an entire day just to establish the correct environment.
These inconsistencies compounded over time, creating a precarious situation where individual components might work perfectly in isolation but fail when brought together. JTEKT had to evolve its design process.
The Model-Based Software Factory Delivers
JTEKT recognized that incremental improvements wouldn’t solve the underlying problems. It needed to transform its software architecture and development processes.
The first major change involved transitioning to a standard architecture. JTEKT developed and implemented a set of modeling guidelines for Simulink® and Stateflow®, breaking down its systems into smaller, more manageable units.
“Model-Based Design with Simulink is a perfect fit with our model-based software factory.”
“The main thing was to rethink our software architecture to make it more structured,” says Bertotto. “We needed smaller units—one component for each function—that would be easy to test, maintain, and understand.”
This shift had profound implications for safety certification. According to Bertotto, “Software complexity dropped. The smaller components were easier to test and validate for compliance with safety standards.”
The modular approach introduced new challenges because smaller components meant more integration work. But this was addressable through automation. The team also shifted from project-specific repositories containing 150 software components to individual Git™ repositories for each component, simplifying component reuse across projects. With the architectural foundation in place, JTEKT built its model-based software factory around comprehensive automation. The centerpiece is a continuous integration pipeline that transforms the entire development workflow.
“Model-Based Design with Simulink is a perfect fit with our model-based software factory,” says Bertotto.
The pipeline, implemented through Jenkins® integration with MATLAB® and Simulink, executes in approximately 30 minutes at the unit level. This includes everything from generating code to creating a simulation model that is ready to be integrated into the global market. When a developer commits changes, the system automatically checks the model against coding standards using Simulink Check™, ensuring ISO 26262 compliance from the start.
Next, simulations verify functionality, generate production-ready C code through Embedded Coder®, and execute comprehensive unit tests. Finally, all required documentation and reports are generated, and Polyspace Code Prover™ evaluates quality metrics.
To support this workflow, JTEKT developed several custom MATLAB toolboxes. “We had our mini-MATLAB marketplace for JTEKT,” says Malrait. The company has also begun deploying the Process Advisor app in Simulink to standardize development tasks.
The system has proven its worth. In 2023, JTEKT executed approximately 3,200 successful pipeline runs. By 2024, that number grew to nearly 5,000, representing thousands of feature updates and bug fixes delivered with confidence.
Impressive Results
The impact shows up everywhere. When a team needed to add monitoring functionality across 40 software components, the automated system saved them 1 hour of effort per component—five person-days total for what would have been a tedious and long manual process.
Fewer resources are required to tackle complex projects now compared to several years ago. According to Bertotto, one current project—developing a steer-by-wire system with entirely new technology and features—runs with 50% fewer engineers on the application layer compared to just a few years ago.
“Using Polyspace Access reduced review time by 50% and streamlined storage of code analysis results. This improved data search time by approximately 15%.”
Development speed has accelerated even more dramatically. Processes that once consumed days are now complete in hours or even minutes. Rapid prototyping, whether on testbenches or in actual vehicles, has become remarkably agile. “Within the day, we can implement, integrate, do some tests, and then ask the driver to test the solution,” says Bertotto.
The automated approach also improved JTEKT’s ability to leverage global resources. The company now successfully externalizes unit testing activities and implementation to its service center in Romania—something Bertotto said would have been “a nightmare” under the old system.
Software component development process.
Quality improvements have been equally significant. The systematic generation of artifacts ensures consistency between models, code, and documentation. Quality engineers and safety auditors now receive reliable, automatically generated reports rather than manually created documents of questionable provenance. These improvements contributed to JTEKT achieving ASPICE Level 2 certification in 2021, demonstrating the maturity of its software development processes.
Perhaps the most challenging aspect of JTEKT’s transformation wasn’t technical but cultural. Despite having a working prototype of the automated pipeline by late 2018, widespread adoption didn’t occur until 2020. Engineers who had spent years perfecting personal workflows were reluctant to abandon them for a centralized system.
“They did not trust it,” says Malrait. “They trusted their scripts, their way of doing things.”
The breakthrough came through a combination of communication, demonstration, and management support. According to Malrait, the management team recognized the time savings potential and incorporated adoption into developers’ yearly objectives. Once adoption began, it quickly became self-reinforcing. Engineers who experienced the benefits became advocates for the system. Today, the cultural shift is complete.
“I don’t think anyone would go back now,” says Malrait. “Jenkins has become so critical to our work that we invest in our computer systems to eliminate the possibility of downtime.”
Building on the Software Factory Foundation
The success of JTEKT’s model-based software factory in transforming steering system development caught the attention of other divisions within the company. The bike electrification project, part of JTEKT’s diversification strategy, provided an ideal opportunity to validate the approach beyond its original domain.
“Support from MathWorks to standardize and unify software engineering development processes, methods, and tools has been crucial.”
Unlike the steering systems team, which gradually evolved over several years, the bike electrification team adopted the software factory approach from the outset. According to Bertotto, the team’s rapid adoption demonstrated natural integration of the new processes.
According to Malrait, the bike team could leverage existing automation tasks, including model checking routines and code generation templates that weren’t specifically built for steering projects. By using Docker® containers for their Jenkins agents, they made the system even more scalable and maintainable.
This enhanced workflow has also been implemented by the Electric Power Steering team in Japan. According to Mr. Keisuke Nagashima, a software engineer at JTEKT, the new workflow has enabled the team to improve their code verification. “Using Polyspace Access™ reduced review time by 50% and streamlined storage of code analysis results,” he said. “This improved data search time by approximately 15%.”
JTEKT continues to enhance its model-based software factory. Deploying Process Advisor across all teams will standardize workflows even further. Containerizing the entire toolchain promises improved scalability and maintenance, with the bike electrification team’s success with Docker containers providing a template for broader implementation.
JTEKT is enhancing its monitoring capabilities on two fronts. System observability tools will track Jenkins’ performance, MATLAB startup times, and other metrics to prevent development slowdowns. Meanwhile, quality dashboards provide real-time visibility into component quality, putting issues directly in front of developers.
“The closer you get the information to the developer, the more likely they will fix it,” says Malrait.
Even with ambitious goals, the engineers who spent years struggling with monolithic code and manual testing can barely believe what they’ve already achieved.
Build results for a software platform during a one-year period, showing the distribution of successful, failed, unstable, and other builds, along with the total and average duration of successful builds.
“If we were asked to do that a few years ago, I don’t think we would have imagined that this is possible,” says Bertotto.
According to Mr. Sakata, “Support from MathWorks to standardize and unify software engineering development processes, methods, and tools has been crucial.”
Read Other Stories
The Orion Spacecraft Is Headed to the Moon, with Help from the SLS Rocket
NASA’s Artemis Program Targets a Long-Term Lunar Presence
Applying Model-Based Design for Automotive Service-Oriented Architectures
Renault Advances ADAS Prototyping with ROS Toolbox and Simulink Integration
Tata Motors Accelerates Vehicle Diagnostics with LLM-Based Chat Assistant
ServiceSage App Boosts Service Speed and Accuracy
