Paving the fast lane to race car optimization with insights from Sibros Deep Logger and Amazon SageMaker

Auto-racing teams continue to push the boundaries of track day performances for both vehicles and drivers. The combination of cross-functional teams involved in driver success goes beyond what is shown during the race event and extends to deep collaborations with automakers. Automakers are engaged in constant feedback loops to optimize the performance of their cars and give racing teams the best possible edge.

Automakers must be intentional about the data collected from the vehicle to help drive success. However, another critical component is effectively analyzing that data to provide meaningful insights. By using machine learning (ML) techniques and tools, it is now possible for automakers to uncover novel insights. In this blog, we show how automakers can use Sibros Deep Logger and Amazon SageMaker to help build, train, and deploy ML models to optimize their vehicle fleets.

To log or not to log: data collection at the edge

Data collection by automakers is a critical pillar for enhancing their fleet performance. Collecting data at the edge while modifying collection parameters on the fly is a powerful mechanism. It helps reduce unnecessary data volume while helping teams at automakers focus on relevant data only.

Sibros Deep Logger is a software product that helps automakers with the logging of any vehicle signal at scale. Sibros Deep Logger has several implementation options available, from pre-integration with partner hardware solutions to a standalone solution that can be deployed on the automaker’s existing infrastructure. The solution supports two types of events for collection. The first is known as “Selective Logging,” where a user can identify a target signal or parameter to capture. The second is known as “Event Logging,” where conditions can be expressed in the form of an event moment.

Track day action

A team from Sibros and Amazon Web Services (AWS) attended the Mid-Ohio Sports Car Course on training day to help a select number of race car drivers optimize driver performance. The team worked with a handful of drivers to first configure Sibros Deep Logger within their vehicles and then select several key parameters for collection using the Selective Logging feature. Captured parameters included GPS coordinates, vehicle speed, throttle position, and brake position, among many more. The team could monitor the inputs in real time as the drivers drove laps around the race track and provide initial insights and observations. Automotive engineers could also observe this data through the Sibros dashboard product “Deep Connected Platform”. After a particular data collection threshold was met, the data was sent by Sibros Deep Logger on the vehicle to the Amazon SageMaker ML pipeline for further analysis by the automotive engineers. The overall goal was to analyze driving patterns and help identify areas for race car drivers to improve their competitive edge.

See the video below:

Solution Architecture

Figure 3. High level architecture of Sibros Deep Logger Workflow on AWS from the race car to the Sibros Deep Connected Platform.

The architecture diagram in Figure 3 shows the connectivity between the Sibros Deep Logger agent installed on the race car and the Sibros Deep Connected Platform. The workflow demonstrates how data is securely transmitted to a Machine learning pipeline that uses Amazon SageMaker to help produce forecasts on the ingested data sets. Stakeholders from the automaker are then able to log into the Sibros analytics dashboard to review the results. Picture of the Analytics dashboard can be seen in Figure 4, where we are viewing a subset of data available for a specific VIN number (Vehicle).

Figure 4. Screenshot of Sibros Deep Connected Platform – Deep Logger Analytics Dashboard

Sibros insights powered by AWS

Amazon SageMaker brings together a broad set of capabilities purpose-built for ML. Data ingested into Sibros Deep Logger triggered an ML pipeline that conducted further analysis for the driver, including real-time predictions on performance for future laps. This created optimization opportunities for specific racing contexts that could be reviewed by automotive engineers, such as a speed adjustment on a particular turn, that could be shared with the driver. With a high precision of accuracy, the Sibros team could predict performance indicators on turns in both traffic and no-traffic scenarios. Specifically indicators such as the time to complete a specific turn, entry speed, exit speed, top speed on track overall, and apex speed.

For our use case, Amazon SageMaker notebook instances were used to perform the ML tasks. An Amazon SageMaker notebook instance is a ML compute instance running the Jupyter Notebook App. Amazon SageMaker manages creating the instance and related resources. Jupyter notebooks within the notebook instance are used to help prepare and process data, write code to train models, deploy models to Amazon SageMaker hosting, and test or validate ML models

In order to perform predictions on different metrics of driver performance, Sibros and AWS members at the race track used the “Time Series Forecasting” technique. This technique forecasts future values using historical data and related covariates. Time Series forecasting can be implemented in Amazon Sagemaker by using the Autogluon ML Library. This library can be used to build and develop models that can be used on targeted datasets selected by automotive engineers. Autogluon Tabular has a “TabularPredictor” function which predicts the values of a target column based on the other columns in a tabular dataset.

There are typically three steps during when using “Time Series Forecasting” in an ML workflow on AWS. The first is identifying the dataset that you want to train and making sure your Notebook instance has the appropriate permissions to access the data. The next step is to train the data which will produce a model that can be used against future datasets to make predictions. A developer can fine tune aspects to optimize for their use case. Fine tuning parameters include setting time limits on how long it can take to train a model. After the model is ready, the developer can use the model to make predictions on new datasets received. This model can now be integrated into a pipeline or workflow by automotive engineers, data scientists or developers to perform on demand. A developer can now use business logic or conditions to load the model to be used in a prediction. Once the model is loaded, the developer can then run the prediction on the target dataset as shown in figure 1.

Figure 1. Jupyter Notebook – Speed Track Predictions

By using Amazon Sagemaker, automaker engineering teams can automate data pipelines that allow for continuous outputs. Teams can also eliminate manual effort associated with cleaning and organizing the data ahead of the ML ingestion point. Furthermore, teams are able to continuously improve their ML models and enrich multiple algorithms over time to help improve the accuracy of the predictions.

Performance analysis is the life force of race car success. The combined efforts of Sibros and AWS produced valuable insights for race car teams to help take their output to the next level. This included key findings on corner performance, exit speed, and competitive analysis as described below.

Figure 2. Graphical view of performance analysis on turn 5 of the Mid-Ohio Sports Car Course and prediction over four race car laps

Corner performance

When diving deep into the analysis of corner performance, it’s important to take a look at the components that make up a corner in the race car context. It is generally accepted that there are at least five parts to consider: entry, turn in, mid corner, pedal transition, and exit. Using telemetry data sent over through Sibros Deep Logger, we can map the exact point on the race track with the car’s internal signals and outputs. Some of the signals captured were throttle position, engine load, and braking position.

Exit speed

Exit speed is the measure of how fast a vehicle exits the corner or turn. This is typically the area where a driver can achieve major gains in terms of time. This insight provides a detailed viewpoint of the losses and gains on lap times.

Competitive analysis

Comparing performance in both traffic and no-traffic scenarios in lap analysis is critical to improving a driver’s toolbox on the racetrack. By focusing on the strengths and weaknesses, teams can put together strategies based on data to increase their chances of improving finish placement or winning.

Benefits

Below are benefits of automakers utilizing Sibros Deep Logger.

• Dynamic and Codeless: Allows automakers to change or update logging parameters and configurations without pushing new code or “Over-The-Air” (OTA) updates.
• Predictive Insights: Leverage Deep Connected Platform and utilize advanced ML pipelines to Generate instant AI-powered dashboards and predictive insights with natural language queries.
• Live or Offline logging: Log, compress and ingest live data to the cloud or as offline files that can be programmatically retrieved at any time.
• Aggregate Visualization: Automatic signal visualization across the fleet with configurable dashboards and alerts
• Any Source or Signal: Collect signals from any sensor or ECU over multiple protocols – MDF CAN, Ethernet/PCAP, OBD-II.
• On-Change Logging: Efficient data collection triggered by deviation detection to log only relevant signals that require action.

Conclusion

Through the use of Sibros Deep Logger along with Amazon SageMaker, we applied deep analysis and ML models to data collected from the vehicle to help optimize the performance on the race track. However, the opportunities goes beyond the race track, which is only one of many use cases that we can help automakers achieve, adding value to current collected datasets and providing new insights for vehicle fleets.

To participate in an innovation workshop and learn how you can unlock value, please reach out to the following contact: marketing@sibros.tech. You can also visit our Sibros Deep Connected Platform offering on AWS Marketplace to learn how you can get started with our product suite on AWS.