An ML-Fueled Car Testing System
About the company
Ferrari, established in 1939, is a renowned luxury car manufacturer recognized for its commitment to delivering unique, high-quality, and personalized driving experiences with a strong focus on safety. Given their dedication to staying at the forefront of performance and consistently offering new and reliable models, they decided to obtain an advanced and smart testing system.
![[object Object] on the map](https://content.andersenlab.com/common-data/client-maps/Italy/1704295477.svg)
About the project
In pursuit of this goal, the customer selected our company as the vendor for this initiative. Andersen was chosen due to our relevant track-record and knowledge across all relevant domains.
In general, we were entrusted with:
- Elaborating failure models and health indicators;
- Building ML models to train the system on historical data to constantly improve the accuracy and quality of failure predictions;
- Developing a predictive maintenance system to reduce operating costs and equipment downtime.
Challenges
The project aimed to address a critical aspect faced by the customer – i.e. their need for thorough testing rounds when developing new car designs.
Such testing is executed by the customer’s dedicated department, equipped with dyno stands (short for “dynamometer”), where various readings and characteristics are collected, including:
- Power;
- Thrust;
- Torque;
- Rotational speed.
These stands are sophisticated setups that feature numerous sensors to measure acceleration, emissions, and other parameters.
A significant challenge experienced by the customer was caused by the susceptibility of this equipment to malfunctioning issues. Even a single failure in a single sensor could lead to a testing lockdown, hindering development processes. Despite the equipment featuring telemetry to monitor and self-diagnose, including capabilities like a rotating wheel shaft and voltage sensors, this limitation persisted.
Thus, the primary objective was to leverage the collected data to establish a predictive maintenance system for the department responsible for testing new cars. For the customer, that would signify a major step forward, as compared with habitual reactive (post-breakdown) and planned (scheduled at certain intervals) testing programs.
As a result of our involvement, a revolutionary predictive maintenance system has been implemented. This system relies on a self-learning data model utilizing ML. The model analyzes deviations, self-learns over time, and enhances the accuracy of its forecasts with each breakdown. These forecasts will generate recommendations for repairs and replacements before any breakdown occurs, optimizing the efficiency of the testing process.
Solution
Discovery phase activities
The Discovery phase initiated by our experts sought to:
- Establish the criteria and characteristics of the equipment;
- Create catalogs of errors and equipment failures;
- Formulate hypotheses of equipment failures;
- Define model accuracy criteria and data input and output;
- Determine the data and formats for issuing recommendations;
- Describe the target system architecture;
- Prepare the documentation for the future development;
- Determine one critical device for an MVP and then extend the system iteratively.
The Discovery phase itself, which was six weeks long, involved a dedicated PO assigned by the customer, and the following staff members assigned by our company:
- A System and Data Analyst;
- A Delivery Manager;
- An ML engineer and Data Science expert;
- A Solution and Data Architect.
Discovery phase deliverables
Joint efforts made by Ferrari in collaboration with Andersen resulted in obtaining the following high-quality deliverables constituting a viable and promising foundation for the development stage.
- The Vision and Scope document
- A set of documents for future engineering efforts
- Architectural and data vision guidelines
Since this project was envisioned as an ML-driven initiative from the outset, data-related project facets were of particular importance here, including:
- Data structure and flows;
- Data lake and DWH;
- ML/DL aspects, involving a defined data model and processes, based on the rigorous evaluation of the existing data;
- Defined ML/DL criteria for forecast models, ML/DL API model, and relevant RNN/CNN models.
Development phase
Following the completion of the Project Discovery phase, our tech specialists initiated the development of the planned system. In order to ensure timely and budget-friendly progress, we proposed a managed delivery framework. This approach strikes a balance between budget management and flexibility in feature implementation, enabling the estimation of scope with necessary adjustments in response to blockers, insufficient detail, lack of information, and other factors.
This has empowered the customer to capitalize on the following advantages:
- Budget control at every phase;
- Mitigated risks and reduced uncertainties;
- Dynamic prioritization of all phases;
- Flexibility in changing the scope of each iteration through Agile practices and ceremonies.
Project results
The entire development process was smooth and fully transparent for the customer, with regular meetings, submitted reports, and knowledge transfer sessions. As a result, the customer was aware of the development progress of the ML-driven system from the very beginning. This transparency greatly facilitated their efficiency in using it. As of now, while some ML activities are still in progress, the dyno stands solution and its algorithms are already contributing to optimized testing workflows for new car designs.
Thanks to our contribution, the following results have been achieved:
- 100% compliance with budget and time limits;
- 92% accuracy of the ML model;
- 40% less money assigned to testing;
- Idle time significantly reduced.
Let's talk about your IT needs
What happens next?
An expert contacts you after having analyzed your requirements;
If needed, we sign an NDA to ensure the highest privacy level;
We submit a comprehensive project proposal with estimates, timelines, CVs, etc.
Customers who trust us
