Reducing Sensors by Using Data Analytics
In today’s automobiles, there are a number of sensors. This number varies from 15 to 50. Luxury models have many more sensors. It is projected that in the next few years, the number of sensors is likely to go up to 200. However, every additional sensor comes with a price tag. The higher the price tag, the less competitive the vehicle is in the market. On the other hand, sensor reduction at the cost of functionality reduction is not the solution. The right approach is to reduce the sensor cost by implementing the same functionality in software.
It is possible to take over the functionality of a sensor provided we have a good analytical method to predict the output of that sensor as if the sensor were actually present. Such analytical method should be able to predict the output on a real time basis. This puts a constraint on the type of equations one can use. They have to be simple enough to ensure that while solving we do not hamper the real time performance.
The problem we were trying to solve was whether there is a way to reduce nine temperature sensors measuring temperature inside the cabin in a luxury vehicle. We started with conducting experiments by heating and cooling the cabin. Temperature was measured at all nine locations where the sensors were supposed to be placed. Thus, we recorded nine channels of data. Each of these channels was analyzed separately. We developed a mathematical model to predict the temperature at nine locations. These equations were kept as first order equations to ensure quick computation. Thus, we did not make use of the Naïve Stokes equations. Our predicted temperatures were within 1 degree C. The main difference was that our calculations did not require high-end processors and all predictions were made on a real time basis. With this performance, we could reduce the nine sensors to only one.
Data analytics can be effectively used to predict values of the sensed quantity. In one particular case, our method predicted values as good as the predictions from solving complex Naïve Stokes equations. The major benefit was that our calculations did not require high-end processors further controlling the costs and all predictions were made on a real time basis.