Machine learning can effectively solve the problem of automotive sensor performance degradation

As part of the transition from older to newer technologies in modern vehicles, inductive position sensors are designed to replace Hall-effect sensors, a shift that is essentially about better managing issues related to automotive sensor degradation.

For example, Microchip has introduced inductive position sensors for automotive applications such as automotive throttle bodies, transmission gear sensing, electronic power steering, and accelerator pedals. Position measurements need to be unaffected by stray magnetic fields and no external magnetic devices are required.

While engineers want to ensure that the sensor will operate over a wide range of temperatures, they worry about changes in the mechanical structure and degradation of the magnetic properties that could affect accuracy. But inductive position sensors use metal instead of a magnet, which does not degrade over time.

“It’s an important component to look at sensor degradation, whether it occurs inside the IC or outside,” said Mark Smith, senior marketing manager at Microchip. “When it comes to sensor degradation, the primary concern for engineers using inductive position sensors should be the longevity of the PCB.”

This is also important because sensor ICs serving automotive applications increasingly require ASIL certification. Microchip's inductive position sensors, the LX3301A, LX3302A and LX34050, are ASIL-B certified, allowing system designers to detect more than 90% of single-point failures.

Figure 1: The LX3302A inductive position sensor uses a larger EEPROM to facilitate eight calibration points to ensure sensor measurement accuracy.

Sensor degradation management

Currently, the industry is managing sensor degradation issues from scratch to meet ASIL certification. What if this transistor is broken, or that circuit fails? If the sensor output is shorted, what can engineers do? "It's a force majeure and time-consuming." Smith said.

Specific experiments must be performed to check or prove certain numbers, also known as coverage. Automotive engineers can create a fault and ensure it can be detected while using industry-standard reliability charts. "It's a relatively simple system that engineers can work on effectively," Smith added.

Today’s vehicles use approximately 50 position sensors, so the transition from Hall Effect sensors to inductive position sensors is critical in managing automotive sensor degradation. In addition to selecting sensors with materials that are less susceptible to degradation, what else can be considered in effectively managing sensor degradation in vehicles? Smith believes that machine learning is the way forward.

Smith said machine learning models can enable pattern recognition in automotive sensors before they fail. “Automotive engineers can analyze five different sensors and detect system-level failures as well as higher levels of degradation.”

Machine Learning is the Future

The automotive industry is starting to take sensor degradation seriously, but over time, there are ample opportunities to leverage machine learning for degradation-related analysis using some advanced computing techniques. However, the idea of ​​using machine learning to manage vehicle sensor degradation is currently in its infancy and requires greater computing power.

Figure 2: Machine learning, taken up to the sensor level, can be used to create models to measure and mitigate automotive sensor degradation. Source: Mathworks

This approach enables engineers to collect large amounts of data, put it into a machine learning model, and then look for possible degradation. This is what is currently being done in the design of autonomous vehicles (AVs). "Machine learning is emerging at the sensor level, and it can be used to simplify the degradation measurement process and make the diagnostic process more efficient," Smith said.

The study of automotive sensor degradation is a natural fit for machine learning. The fact that machine learning takes large amounts of data and puts it into a model to detect sensor failures can lead to significant improvements in reliability and cost savings.