Machine learning and automotive tuning: Seletron's know-how
What is "machine learning," and how can this technology be applied in some way to the car tuning sector?
The term machine learning outlines the ability of some machines (more aptly computers) to collect data and information and process it for self-management and improvement, somewhat like what happens in our brains. There are different levels of machine learning, and they are branches of artificial intelligence. These are methods of analysis that can collect a large amount of information and recognize patterns to improve the performance of the system itself.
There are many uses for this branch of science as the system enables the progressive improvement of adaptive software capabilities and the reduction (or potential nullification in some cases) of the need for human intervention. Machine learning algorithms learn information (careful with the word "learn," as it can sound rather creepy...) by acquiring and processing data, thereby modifying their own behavior.
This is a form of artificial intelligence where the system is not created with all the desired behavior already finalized but with a learning program, and the machine autonomously (or partially autonomously) develops its capacity to collect and analyze data, becoming more "intelligent." In some cases, the collection of data and its subsequent processing are separate from one another (like with semi-auto-driving cars that can collect a mass of data, send it to a remote, centralized processing center, then receive the product of the processing in the form of a software update that is the result of learning), in other cases, the machine (computer) has both the elements necessary for data collection and the artificial intelligence necessary for its processing. The result is the implementation of both the autonomy and performance level of the process that the computer oversees.
How can this branch of artificial intelligence be applied to the world of automotive tuning? SELETRON's research and development department is introducing some features that stem from the basic concept of machine learning. To understand how this approach can impact electronics for engine tuning or electronics tasked with simulating the presence of catalytic converters or particulate filters (DPFs, FAPs, or OPFs, whatever they may be), we need to begin by looking at the "traditional" process of developing these kinds of projects.
In the normal development phase, we start with the need (or problem to be solved) by devising a possible solution based on previous data collection, e.g., data regarding the control process of the electronic system to be modified, data from real operating values, etc. A subsequent stage involves the creation of prototypes that are then installed on several cars constituting the test vehicles. Adjustments are then made to obtain the desired results, and when the operation appears perfect under all normal engine operating conditions, they are tested at length to confirm the absence of problems that may occur in the presence of specific variables (e.g., very low/high outside temperature, fuel type, engine temperature, miles driven, atmospheric pressure, driving conditions, etc.). Only then is the product engineered to optimize various aspects of the product so that the tuning modules can be mass-produced.
The approach that embraces the philosophy of machine learning involves the design of more complex electronic control units capable of autonomously collecting various data during normal engine operation (for catalytic converters or particulate filter presence simulators, this phase takes place BEFORE the mechanical modifications) to recognize patterns that will then be digitally recreated to "trick" the original ECU into believing that the element removed from the exhaust system (filter or catalytic converter) is still present. This is becoming increasingly important as the electronics installed on modern engine control systems and related devices are also becoming more and more complex.
To rephrase, whereas "traditional" chip tuning modules have a system that is programmed with predefined software (which already contains the algorithms that are the result of previous data collection and subsequent processing by SELETRON technicians), chip tuners designed taking into account the principles of machine learning starts by installing electronics that can FIRST acquire a mass of data, and THEN process it to adapt, improve, and control sensor signals to be consistent with what the car's stock ECU "expects." This is obviously a simplification of a broader concept, but the topic demonstrates once again how SELETRON seeks to be not only "in step" with the best tuning technologies and solutions but wants to advance them by setting itself up as a pioneer in the industry
If you want to see the options for chip your car click the catalog in the button below