ML uses pattern recognition to perform sensor-degradation analysis
Article By : Majeed Ahmad
While the automotive industry is looking at the sensor degradation problems deterministically, machine learning offers a venue to perform degradation-related analysis using pattern recognition models.
As part of modern vehicles’ transition from older technologies to new technologies, inductive position sensors aim to replace Hall effect sensors, and this transition is intrinsically tied to better management of issues related to automotive sensor degradation.
For instance, Microchip Technology has unveiled inductive position sensors for automotive applications such as automobile throttle body, transmission gear sensing, electronic power steering, and accelerator pedals. The value proposition: position measurements are immune to stray magnetic fields and don’t require an external magnetic device.
While automotive engineers want to ensure that sensors work across a range of temperatures, they are concerned about variations in the mechanical structure and magnet degradation, impacting the accuracy. On the other hand, an inductive position sensor uses a piece of metal instead of a magnet, and the piece of metal doesn’t age much over time.
“That’s a big component to watch in sensor degradation, whether something happens to IC or externally,” said Mark Smith, senior marketing manager at Microchip. When it comes to sensor degradation, engineers have to mostly worry about the longevity of the PCB when using inductive position sensors, Smith added.
It’s also crucial because sensor ICs serving automotive applications increasingly require ASIL certifications. Microchip’s inductive position sensors—LX3301A, LX3302A, and LX34050—comply with ASIL-B certification, allowing system designers to detect ≥90% of all single-point failures.
Figure 1 A greater EEPROM space in LX3302A inductive position sensor facilitates eight calibration points to ensure sensor measurement accuracy. Source: Microchip
Sensor degradation management
Currently, the industry is managing sensor degradation-related problems from the ground up to comply with ASIL certifications. What happens if this transistor fails or that circuit malfunctions? What can engineers do if a sensor is short on output? “It’s a very deterministic and time-consuming approach,” Smith said.
Specific experiments must be carried out to check or justify certain numbers, also known as coverage rates. Automotive engineers can create a fault and ensure that it can be detected while using reliability charts from industry standards. “It’s a relatively simple system, and engineers can efficiently handle them,” Smith added.
Today’s vehicles use around 50 position sensors, so a shift from Hall effect sensors to inductive position sensors can be critical in managing automotive sensor degradation. Beyond the selection of sensors in which materials don’t age much, what else is on the cards in efficiently managing sensor degradation in vehicles? Smith believes machine learning is the way forward.
Smith said that machine learning models could implement pattern recognition before failures show up in automotive sensors. “Automotive engineers can analyze five different sensors and detect a system-level failure as well as degradation at a higher level.”
A new book, AspenCore Guide to Sensors in Automotive: Making Cars See and Think Ahead, will help you make sense of the sensor labyrinth in modern vehicles. It’s available now at the EE Times bookstore.
Machine learning is the future
While the automotive industry is looking at the sensor degradation problems very deterministically, moving forward, there is an ample opportunity for using some of the advanced computing techniques to perform degradation-related analysis using machine learning. However, the idea of using machine learning to manage sensor degradation in vehicles is currently in infancy and will require far more compute power.
Figure 2 Machine learning, up and coming to the sensor level, can be used to create models for measuring and mitigating automotive sensor degradation. Source: Mathworks
This approach enables engineers to collect a bunch of data, put it into a machine learning model, and then look for a signature. That’s what the autonomous vehicle (AV) designs are doing right now. “Machine learning is up and coming at the sensor level, and it can be used to simplify the degradation measurement process and make the mitigation process more efficient,” said Smith.
Automotive sensor degradation marks another venue where machine learning has an opportunity to win. The fact that machine learning takes a lot of data and put it into a model to detect sensor malfunctions can lead to substantial reliability gains and cost savings.
This article was originally published on EDN.
Majeed Ahmad, Editor-in-Chief of EDN, has covered the electronics design industry for more than two decades.
Related articles:
- 3 basic facts about automotive sensor degradation
- Book review: A guide to sensors in automotive
- When can a sensor be used in an automotive system?
- Startup Accelerates AI at the Sensor
- Deep Learning on MCUs is the Future of Edge Computing