Develop to Survive.

“Millions of people have already watched the series ‘Drive to Survive’ on Netflix, which shows the behind-the-scenes of the world’s premier Motorsport category, Formula 1. However, there is also a backstage that has not yet been revealed to motorsport fans — the race for software and simulation algorithm development within each team.” — Fábio Mori, author of the article.

In a universe where victories are won in milliseconds, every decision is a crucial piece in the journey towards success. With exponential advancements in artificial intelligence and machine learning, software has evolved, absorbing every detail recorded by drivers during their journeys on the tracks. But can these artificial intelligence models truly join the engineering team in pursuit of ultimate performance? In this article, we unravel the development journey of an innovative software applied to Motorsport, through graphical visualizations, automated tasks, and, of course, AI-informed decision-making.

Stock Car

The biggest and most traditional Motorsport category in Brazil, it is a true spectacle on wheels that enchants motorsport enthusiasts across the country. With its thrilling races, the category has won thousands of passionate fans, turning each event into an authentic celebration and filling host cities with excitement and energy.

“Stock Car was created by the Brazilian Chevrolet Dealers Association, inspired by the American NASCAR. The first race took place on April 22, 1979, at the Tarum? racetrack in Viam?o-RS.” Official Stock Car website.

Besides being an entertainment phenomenon, Stock Car plays a fundamental role in the economic sphere. Research data indicates that the category’s events have a significant impact on the Brazilian economy, generating direct and indirect employment opportunities in various areas. From the preparation and organization of races to the tourism, hospitality, and commerce sectors, the financial movement is noteworthy.

“The success of Stock Car on racetracks and its exposure on TV began to attract top-notch drivers, with experiences in the world’s leading motorsport categories, like Rubens Barrichello, who won the title in only his second season in the category.” Official Stock Car website.

The influence of Stock Car goes beyond the tracks and teams. Sponsors actively compete for the opportunity to link their brands to this category full of glamour and emotion. The competitive scenario and the high visibility of the races make Stock Car a powerful platform for major companies to strengthen their presence in the market and connect with the audience.

“Currently, 38 million Brazilians declare themselves fans of the category, representing 35% of the connected population.” Sponsorlink — The world’s largest specialized sports research.

In the highly competitive racing environment, each team constantly seeks strategic advantages and technological innovation. Motorsport categories around the world are considered true laboratories of new technologies due to the need to find new solutions for an environment of extreme competitiveness.

“Stock Car crowned a total of 15 different winners in the 24 races of this season — practically one winner for every two drivers on the grid. Six of them stood on the top step of the podium for the first time in the category. The qualifying session had 22 cars within half a second and 30 within one second, in the 11th round in Goiania. The Toyota Corolla model, the champion, won 10 races. Its competitor, the Chevrolet Cruze, triumphed in 14 races, almost evenly dividing the victories up for grabs. In total, 27 drivers led a race in the 2022 Stock Car season.” — Flavio Souza, Torcedores.com.

Thus, Stock Car consolidates itself as much more than a sports category, it is an event that combines passion, technical performance, and economic impetus to boost Motorsport in Brazil and inspire generations of fans and drivers to accelerate towards success.

“Motorsport has become the sport of statistics. It is in its DNA. Motor racing is measured in numbers: maneuvers are transformed into milliseconds, victories and poles generate rankings of memorable achievements. Not surprisingly, the world’s premier categories possess elaborate databases, results of detailed research to record decades of history, records, and statistics.” — Fernando Julianelli, CEO of Stock Car.

How Artificial Intelligence Can Revolutionize Motorsport.

Behind the powerful engines and fast curves, teams’ computers are at the epicenter of a technological revolution that has changed the way racing is conducted. Hundreds of strategically installed sensors in the cars generate a flood of data that flows into the teams’ laboratories. The ability to make the most of all this information is crucial for survival in a highly competitive environment.

“There are over 300 sensors in the cars, which allow the team to take advantage of all kinds of information, including external data,” explains Geoff Willis, the technology director of the Mercedes team.

In addition to the sensors installed in the car, teams collect a lot of external data, such as timing and weather data, and perform feature engineering through mathematical channels that generate new information based on the acquired signals.

“We generate between 5 and 10 terabytes of data per week. It’s a lot of data, and managing it is a big challenge, but it offers us many opportunities,” adds the Mercedes technology director.

All the data needs to be processed and analyzed by a specialized team, providing a series of insights that aid in car development and the strategies used during the race.

“These technologies enable a quick and interactive visual analysis of pre-event simulations, providing results that help us make real-time decisions. We heavily rely on our experts and their skills,” says Willis.

In addition to visualization tools and graphical analyses, with the advancement of technology, it is now possible to use this database to make future lap time predictions and setup parameters, further assisting the team in their decision-making.

“You have the data, insights about it, a forecast for the future, and decision-making, which generates learning and feeds back into the system,” explains Márcio Arbex, Pre-Sales Director of Tibco in Latin America.

In the global Formula 1 scenario, investment in technology is a growing phenomenon. Industry giants like Oracle in Red Bull, AMD in Mercedes, and AWS in Ferrari recognize F1 as an innovation laboratory, driving the development of new technologies that may eventually find applications in other areas.

The partnership between the Red Bull team and Oracle allowed the optimization of data usage through the Oracle Cloud Infrastructure (OCI), boosting race simulations throughout the season to improve car settings and provide detailed information for strategic decisions during the races. During the 2021 championship, Red Bull increased the number of simulations executed with OCI by 1,000 times, enhancing the accuracy of predictions and decision-making. Additionally, they increased the simulation speed by 10 times, giving race strategists more time for assertive decisions and demonstrating how the integration of advanced technologies and efficient data analysis drive Motorsport to new levels of competitiveness and excellence.

In 2022, Mercedes innovated by adopting 2nd generation AMD EPYC? processors, resulting in a remarkable performance increase of about 20% compared to previous servers. This strategic change allowed the team to perform twice as many aerodynamic modeling iterations daily, significantly accelerating the development of race cars and driving more competitive projects.

“Highly successful Formula 1 teams use AMD technology to design the car, analyze data, run simulations, and, therefore, have this brand association effect,” says John Taylor, CMO of AMD.

AWS in Ferrari: Ferrari will leverage AWS’s advanced analytics, machine learning, computing, storage, and database capabilities to quickly obtain information about the car’s design and performance on the track.

“We chose AWS because of its constant pursuit of innovation, the wide range of machine learning solutions, and its proven experience in supporting partners on a global scale,” says Mattia Binotto, director of Ferrari.

And how is the development landscape in Motorsport in Brazil’s Stock Car? The answer is clear. The same appeal for technological development and strategic use of data to gain a competitive advantage is pulsing here too. Increasingly, Stock Car teams are adopting similar approaches, optimizing data analysis to improve the performance of their cars and drivers.

A testament to the need for development to survive and remain competitive in the coming years is approaching. Looking to the next season, the new Stock Car car for 2024 promises exciting innovations with telemetry systems, providing real-time data for the teams. In this environment of constant evolution, the team that best interprets and uses this data to its benefit will win the race for excellence and stay one step ahead of the competition.

Purpose of the Software.

This article aims to present the development of software for data analysis, report generation, and application of artificial intelligence, with a focus on optimizing decision-making in the Motorsport scenario.

“The greatest challenge in software development lies not only in the thousands of lines of code but also in standardizing tools, file naming, and, most importantly, knowing what information we want to extract from the data. Business knowledge is essential before creating the tool.” — Fábio Mori, author of the article.

Database.

Four sources of databases were explored to enhance performance and decision-making in the Motorsport universe, namely: timing data, car sensors, tire wear measurements, and setup parameters.

The first data source comes from the official timing system of Stock Car, available on the Chronon website. Upon accessing the website, we obtained files in .csv format containing information such as lap times for each driver and their respective speeds.

The second data source is generated by the sensors installed in the cars. These sensors collect a vast amount of real-time information about the behavior of the vehicles during the races. Using the Pi Toolbox software, we exported this data to .csv files, carefully selecting vital data related to the car’s performance.

“A significant amount of feature engineering is done through the car sensors. For example, using only the accelerometer, speed, steering wheel, and pedals data, we can create a large number of mathematical channels that represent important performance-related values, all of which are exported in this file.” — Fábio Mori, author of the article.

The third data source comes from tire measurement spreadsheets, where mechanics meticulously record their wear during practice and races. This data is essential for the precise analysis of tire behavior and for strategic decision-making related to the appropriate choice and use of tires.

Finally, the setup database comprises vehicle parameters that can be modified by the engineering team before and during each practice and race session. These strategic adjustments allow optimizing the car’s performance and adapting it to various track conditions.

These data form the foundation of performance in Motorsport, and accurate analysis of each piece of information is crucial for the decision-making and strategies used by teams and drivers to win races.

Beyond the Numbers: EDA, Visualization, and Feature Engineering.

“It is important to make it clear that all the data used in the images of this article was randomly generated for this purpose and does not represent actual data from any team. This was done to maintain the confidentiality of the information.” — Fábio Mori, author of the article.

Exploring timing data.

Dealing with an external data source, where we have no control over the information’s standard or format, makes the challenge even greater. In Stock Car, unlike Formula 1, where an API provides real-time data, we had to adopt an alternative approach to obtain these valuable insights. Currently, data extraction is performed from .csv files made available after each race session.

These files contain data such as sector times, total lap time, speed, time, and position of each driver in a session. The developed algorithm automatically extracts all this information from all files saved in the project’s data folder structure, creating a historical database used by the software.

In addition to the original information, the algorithm creates new attributes for a more in-depth analysis. For example, with the time of each lap of the driver in hand, we can calculate the “Gap” variable, which represents the time interval between the moment the car crossed the finish line relative to the car ahead. Following this approach, we also calculate the “Gap to Leader,” which measures the time difference between a car and the race winner over the laps.

To further optimize the development of machine learning models, we create variables that act as targets in the models. Regarding qualifying times, we add the cutoff times for each session to the lines of each lap, as well as information on whether the driver advanced to Q2 (Top 15) or Q3 (Top 6). After all processing, the algorithm creates visualization screens for data analysis with charts created using Python’s Plotly library.

Sector times 1, 2, and 3, as well as lap time, are plotted against the number of laps or time, an interesting option when observing differences between groups during practice.

Speed analysis is another very important factor, and when starting from a Python database, it allows us to group this value in various ways before visualizing it. For example, besides analyzing by car, we can examine speeds between manufacturers, in the case of Stock Car, Toyota, and Chevrolet.

Or compare speeds between teams composing the grid of the category.

To find correlations between data, the software has scatter plots that combine created information, such as “Gap,” with speed values, to try to understand if there is a correlation between the distance between cars, generating an effect we call “draft,” which influences the speed of the car behind.

Tabular visualization is also essential when we want an overview of session times, seeing differences in each sector and the driver’s ranking. Filters to select only cars from a specific manufacturer, or all cars, are another important tool during analysis.

Another piece of information created from the feature engineering applied to the database is the “Gap to leader,” which calculates the distance to the race leader, providing a more detailed analysis of performance and strategy during races.

With the normalized database, we can analyze the percentage differences between sectors, speed, and lap time, identifying the car’s strengths and weaknesses in each section of the track.

In addition to creating a powerful Data Analytics tool, the software provides task automation by generating automatic reports based on the data from each session. The importance of this automation lies in saving time and speeding up teamwork, replacing the time spent on creating reports that required more manual involvement with an algorithm that produces customized and automatic information, saving files in .pdf format in folders accessible to the analysts.

“Replacing patterns with scripts is one of the great advantages of this technology. Now, the time that was used to prepare a report is used to analyze the data generated by it.” — Fábio Mori, author of the article.

The sensors that ensure the car’s functioning.

The data generated by the data acquisition and visualization software, in the case of Stock Car, the Pi Toolbox, has a much more predictable pattern than timing data. Essentially, the script automates the concatenation of files from all stages and seasons into a single data source to be used by the software. The implemented procedure of saving these contents with predefined folders and naming ensures that the final file can be joined and organized for use.

“Differently from timing data, which is available for all cars in the category, the data from the acquisition module in Stock Car is the property of each team, so each one obtains this information from their own cars.” — Fábio Mori, author of the article.

Car sensors can be divided into two categories: vital data and performance data. Let’s first talk about the vital data; it comprises all information related to the correct functioning of the car, and its values must operate within a known range.

Examples of these signals are temperatures, pressures, and even the battery voltage level. In the software, it is possible to analyze both the average and maximum temperature values, which may represent some signal spikes during the session.

Another interesting way to compare the operating ranges of each vital signal is through the Box Plot graph, which clearly shows the operating intervals of each signal.

Tabulating this information also allows for the automatic creation of a report where, after each session, the head mechanic can quickly and graphically see if the car’s vital data operated correctly. This is possible because we can color each interval according to the expected operating values, as shown in the example below, with values in green within the safe operating range, in red outside, and in yellow representing a small interval between the two.

Extraction of performance metrics.

In addition to sensors representing vital data, there are signals such as steering wheel, accelerator and brake pedals, wheel speeds, lateral and longitudinal accelerations, among others, that can be used to extract performance metrics from the car.

As these are also data originating from the acquisition and analysis software Pi Toolbox, the feature engineering process through mathematical calculations is done within it, and the exported data already represent these values, or Key Performance Indicators (KPIs) of performance, as they are known.

Therefore, the software again concatenates all these files, saved in folders and with predefined names, to create another database that will be used for visualization, analysis, and report generation.

Some examples of performance KPIs are the Grip Factors, which represent the lateral and longitudinal forces of a car in specific sectors of the track, such as braking, mid-corner, and acceleration (traction), for example.

Mathematical calculations like derivatives and integrals can also generate important indicators about the car’s behavior. In Motorsport, we analyze two main types of behavior: oversteer and understeer. Both can be characterized through the signal of the steering wheel. The first characteristic has a larger variation than the second, so the value of its derivative should be higher. When we evaluate the integral, a car with understeer behavior should present higher values than a car with oversteer, considering that a larger steering angle is needed to make the turn.

In addition to visualizing the mathematical channels, having control of the database within the software allows us to create other types of visualization, such as the correlation between variables. This allows us to analyze whether there are values or intervals where controllable parameters, such as brake balance, which can be adjusted by the driver, represent higher calculated efficiency indexes.

Statistical tools like the Pearson correlation can also be applied to understand whether there is a strong or weak, positive or negative correlation between a chosen target variable. For example, it is possible to understand which Grip Factor has a stronger correlation with lap time on this track.

In addition to these visualizations and others about software performance, we can generate reports with the chosen performance KPIs to have a concise and comparative documentation of the team’s cars, replacing the time that was previously spent on report generation with analyzing these values now.

Impact of tire wear on performance.

It’s not only highly technological sensors installed in cars that provide important information for a Motorsport team. When we look at the entire ecosystem within a racing team, we can find several sources of data that are fundamental for analysis, understanding, and decision-making.

During an event, where there are various practice, qualifying, and race sessions, the mechanics of each team make measurements on the tires after the sessions, representing the amount of rubber present on the inner, intermediate, and outer parts of each tire. These measurements help the engineers understand tire wear and define setup adjustments correlated to it.

This data source comes from the operation spreadsheets of the mechanics at the track, where, once again, standardization of file and folder naming is necessary so that a script can automate the process of concatenating these values into a single historical database to be used by the software.

“When we talk about creating a data project, the first and most important step is to identify what sources are available for the project. Often, we have various data at our disposal that we are not using, and to identify them, business knowledge is fundamental.” — Fábio Mori, author of the article.

The Plotly library, used in this software, provides a series of possibilities to create charts in various forms that represent what we want to analyze. In the case of tires, we represent the wear measurements in a way that allows us to “see” how the tire is.

The control of information in the database allows for not only individual analysis of each car but also direct comparison of how each tire is for each Set (a set of 4 tires).

With the tire rubber measurements, we can perform feature engineering to represent other important information. By comparing these measurements, it is possible to understand not only the current state of the tire but also how the wear occurred in each session or between consecutive measurements. Additionally, correlating this wear with setup parameters such as camber, pressure, and temperature assists the engineer in understanding this analysis.

The choice of tires for the race is also an important factor that requires careful analysis by the engineers. In Stock Car, it is mandatory to change 1 tire (as of the 2023 regulations) in each Pit Stop of each race. Moreover, the strategy of each car in each of the two races that make up a stage in Stock Car will also define the number of tires to be changed.

From a variable called Tyre Grip, it is possible to create a Rank to guide the engineers’ tire choice. This variable correlates information not only from the tire measurements but also from the tire’s mileage and number of cycles; the higher the value, the better the condition of the tire.

In defining the race strategy, various scenarios are considered and decided upon before the start. To facilitate this visualization, an interactive environment with colors for tires that will start the race in green, those in strategy 1 in yellow, strategy 2 in purple, and strategy 3 in gray, facilitates not only the engineers’ decisions but also the understanding of the mechanics, as a document is generated with this information.

In addition to visualization and before using machine learning tools, it is also possible to create adjustment tools for values based on mathematical calculations defined by the engineers. One example is the tire pressure adjustment tool according to track and tire temperatures. In Stock Car (at least until 2023), there are no internal tire pressure and temperature sensors, so the control and adjustment are based on references and cold (pre-usage) and hot (when the tires arrive in the pits) measurements.

The important thing is to understand how just 4 measurements per tire at the end of each day of practice sessions can generate a large amount of information, visualizations, and documents that not only assist the team’s decision-making but also facilitate the work and communication of this information.

Setup parameters, in search of the best performance.

The last database used in this software is the setup parameters. In Motorsport, it is one of the most complex areas of analysis and decision-making. Millimeters in wheel alignment adjustments, axle heights, adjustments in degrees of camber, caster angles, and many other factors that influence the car’s behavior are meticulously defined by the engineering team. There are several simulation software based on vehicle dynamics equations that try to determine the best combination of setup parameters and analyze their influences on the car’s behavior.

The chapter on using machine learning for setup simulation is further ahead in this article, but for now, let’s show how, once again using a tool utilized by mechanics, we can create visualization, analysis, and automatic report generation within the software.

The workbook files, used for annotating times and other parameters during a practice session, such as tire temperature and fuel quantity, must be saved following a predefined naming and folder pattern so that the script can automate the generation of a historical database of this information. These spreadsheets also contain setup parameter information recorded by the mechanics before and after each session, as well as adjustments made by the engineers during a session.

A comparative visualization before and after each practice session, as well as how the setup was during each car’s outing in the session, aids in the analysis of performance data, including driving performance, as the car + driver combination should always be considered as a whole in these comparisons.

For decision-making regarding the setup, analyzing the development and changes made to a car during a stage is also important to understand the line of thinking used by the engineering team and to assess which tests were successful or not.

As important as understanding the best combination of setup parameters is understanding the influence of each parameter on the target variable, which in general is always the lap time. Assuming linear correlations between them, the Pearson correlation can once again show the correlations, positive or negative, of each parameter with the lap time.

“The software has a powerful Data Analytics system, as it can correlate different databases, increasing the chances of the user extracting information and arriving at the correct conclusions. Moreover, all databases are integrated into the same tool.” — Fábio Mori, author of the article.

Machine learning with the data.

Consolidating so much data into the same software goes beyond being an analysis or automatic report generation tool; it opens up an important possibility of allowing the machine to learn from the data. Lap times, vital car sensor signals, performance sensor signals, tire information, and setup parameters — all of this can be used for machine learning, but what needs to be learned is the crucial aspect.

Once again, business knowledge is essential to understand the answers we want to arrive at. First, let’s talk about lap times.

Qualifying cutoff times and success probabilities.

Before explaining this model, it is important to clarify that, according to the regulations in place for Stock Car (in this case, in 2023), the qualifying session, which determines the starting grid order for the race, is composed of 3 distinct sessions. “Q1” is the first part of the qualifying session, where all cars take to the track to set their best lap times. The top 15 lap times advance to the second session, Q2. Finally, only the top 6 lap times from Q2 qualify for the last session, Q3. In Q3, the drivers compete for the best time that will define the first 6 positions on the grid.

When we created the “Top 15” and “Top 6” variables during the data processing phase for lap times, there was an objective behind it — to create a target variable for a logistic regression model. These models classify a dataset into a variable that can take on 2 categorical values, in this case, either “Top 15” or “Not Top 15,” and either “Top 6” or “Not Top 6.”

Being within the “Top 15” means that the lap time ensures qualification for Q2, and being within the “Top 6” guarantees qualification for Q3. If we train a logistic regression model with a historical database of lap time data, we can use this machine learning model to classify, for example, the laps of a practice session and analyze their potential for qualification.

Moreover, if we have a real-time data reading model that reads the lap time data from the screen, we can have this information for each sector, for all cars, calculating the probability of each car advancing to the next qualifying session.

“In Stock Car, completing one less lap in one of the qualifying sessions can be the difference in starting in the first position. It is a very difficult decision to make, to ask the driver to come into the pit before time runs out; you end up taking a very high risk. Having a tool that helps you make this decision when needed can be crucial.” — Fábio Mori, author of the article.

In addition to the “Top 6” and “Top 15” variables, 4 numeric variables were also created, representing the lap times for sectors 1, 2, 3, and the total lap time for the car that finished the session in the 15th position and also in the 6th position. The logic behind this model is the same, but instead of using logistic regression, linear regression is used to predict the cutoff times, along with the times for each sector, to qualify for the Q2 and Q3 sessions.

Analyzing these times before a qualifying session helps understand in which sectors a car is strong, where the biggest differences are, and also the potential of each car, making them important numbers for the team during qualifying.

Predicting tire wear in a race.

After determining which tires will be used in the race strategies, it is necessary to understand if these tires will be able to complete all laps of a race or if an additional tire change will be required, or even which tire position is more critical and therefore needs to be replaced as well.

However, it is not only the number of laps that defines the Tyre Grip target variable, as we have seen before. Camber, tire temperature, and pressure are also factors that affect this wear, in addition to the track, and no less importantly, the driver. All these factors are considered in the calculation of the Tyre Grip variable.

Thus, through a historical database of tires, we train a linear regression model that calculates the value of Tyre Grip based on the input values: track, driver, tire position, camber, pressure, track temperature, and mileage.

“There are tracks with higher abrasiveness that may require us to change more than just the mandatory tire in the Pit Stop. Additionally, during qualifying, we optimize the setup for maximum performance, and sometimes we need to change these parameters, such as camber, to ensure that the tire withstands all laps in a race.” — Fábio Mori, author of the article.

Data-driven setup simulator.

The top teams in the main categories of Motorsport worldwide invest heavily in simulation, as mentioned at the beginning of the article. Aerodynamic simulations and simulations based on vehicle dynamics equations work together to define not only the design parameters in building a car but also to predict the best combination of setup parameters.

This software shares the same objective — to predict the best setup combination that represents the shortest lap time, the target variable in this model. Furthermore, it provides an analysis of influences indicating which parameters have a greater or lesser impact, positively or negatively, on the model’s outcome.

However, the approach here will be different. We will not use any vehicle dynamics equation to predict the result; instead, we will try to reach the answer by taking the opposite approach, starting from the consequence to find the cause, in other words, starting from the data.

The historical database, when created from the workbook tools, contains information on all setup parameters, times for each sector and lap for that configuration, as well as tire, temperature, and fuel information. All of this serves as a learning source for the machine to attempt to predict which parameter configuration yields the shortest lap time.

A linear regression model was trained to learn from this data, and it is important to clarify that all models used up to this point are baselines, meaning simple models to achieve initial results, but they can be replaced by more complex models, even neural networks, as this development progresses.

Within the software, it is possible to define maximum and minimum values for each parameter and also the step of variation for these parameters that you want to simulate.

As a response, it returns the parameters related to the calculated shortest lap time.

And the influence of each setup parameter on the lap time, according to the learning of the machine learning model.

“I am not aware of other setup simulators for Motorsport that follow the logic of data. It is all very new and needs to undergo many tests to be validated, but I believe a lot in its potential.” — Fábio Mori, author of the article.

A cloud software.

For the entire team to have access to the software, it is necessary for it to be shared in the cloud. There are several options available in companies like AWS or Microsoft’s Azure, but there are also simpler paths that can be used in projects like this, where only a few people will have access to the software. GitHub, a versioning and code sharing platform, allows you to create project repositories with up to 100 MB, which may be suitable for a small team and a smaller database like this.

The Streamlit library used in this project to create the entire frontend and visualize the software has a cloud tool called Streamlit Share, where you can share an application developed and shared in a repository in your GitHub account through a link provided by the cloud.

“The cloud allows team members to access the software at any time, with all functionality available at their fingertips, just a click away.” — Fábio Mori, author of the article.

It doesn’t end here.

The development of software is an ongoing process. New needs, new data sources, changes in the information structure, all of these can impact or generate new demands within the software.

In addition, machine learning models must be in constant development, and their metrics should be analyzed to ensure proper functioning.

In any case, there are potential paths that we should research and experiment with. Maziar Raissi, a professor at the University of Colorado and author of various articles on deep learning, has published studies on “Physics Informed Deep Learning — Data-driven solutions and discovery of Nonlinear Partial Differential Equations.” This neural network architecture includes nonlinear partial differential equations in a data-driven approach, allowing us to combine the two sides of the coin, data, and physics.

I strongly believe in the potential of simulating a data-driven setup, but it may have many weaknesses due to the lack of a theoretical and robust foundation that explains all this. On the other hand, software based solely on equations will never learn what the data has to teach, distancing its responses from the reality we see on the tracks. But what if it were possible to combine both sides of the coin? What if it were possible to create a software that learns from a historical database and has the support of vehicle dynamics equations to converge on an optimal solution for adjusting setup parameters?

Perhaps it is no longer a “what if it were possible” but rather a “we must start now,” Develop to Survive.

“The development of software is something with a beginning, a middle, but no end, at least as long as it is still in use. The context of ‘Develop to Survive,’ the title of this article, is not by chance.” — Fábio Mori, author of the article.

About the author.

A Brazilian, 33 years old, an electrical engineer from FEI, and the founder of edutech Escola Matriz. With over 12 years of experience working in electric and combustion-based motorsports, currently in Stock Car. A master’s degree in engineering from UNICAMP with research on the management and control of lithium batteries with machine learning and a postgraduate degree in machine learning and data science from TERA. Holds dual citizenship, Italian and Brazilian. A sports enthusiast, a triathlete, and an ultramarathon runner.

References.

Reports.

Technical articles.