To meet customer’s growing expectations of cabin comfort and surpass the unrelenting benchmarks set by the competition, more and more automobiles today come equipped with power adjustable seats. With this feature in place, engineers are now devising a protective mechanism with which to prevent obstructions. However, making that safety measure a reality still poses difficulties.
Because of the costs involved, the technology in use must dispense with extra sensors and further hardware components wherever possible. In other words, the solution needs to be purely based on software. Existing sensor signals – such as rpm or the adjustment motor’s current – must be made to relay parameters for situations in which an obstruction could occur. These would help pinpoint when a seat is actually obstructed. What makes the development work so complicated? Just some of the many issues include the wide range of power adjustment mechanisms, the different physiques of passengers and their unpredictable movements – not to mention all the kinds of situations in which seats can be obstructed.
One indication that the seat is being obstructed is a marked drop in rpm. This kind of drop is caused by blockage or by a heavy, abrupt and intense load placed on the adjustment motor (since the seat adjustment is moved against the person whose seat is being obstructed). The backrest angle adjustment often uses a mechanism which would cause the signal flow in the adjustment
motor to oscillate quite dramatically during normal operation. The basic problem, however, in identifying obstructions comes down to reliably differentiating the oscillations merely tapering off from the breaks in rpm which are characteristic of obstructions.
Steinbeis professionals at the Steinbeis Transfer Center Automotive Electronics and Mechatronic Systems in Friedrichshafen, Germany are devoting their energies to implementing a software-based obstruction sensor unit for power adjustable seats. Using a wavelet transformation of the rpm signal, these experts characterize the symptoms that accompany seat obstructions. The depiction of the parameters – in other words, making a decision based on symptoms indicating whether or not the seat is being obstructed – is augmented with a neuronal network.
Figure (a) shows a sample rpm progression for an inclination drive under stress. In contrast, figure (b) shows the rpm signal in a situation in which an obstruction might occur: at around 9300 ms the signal’s trajectory drops continuously. This means that the engine is seizing up because an obstruction is occurring. Both test runs were performed with the same load placed on the seat and under 13 V.
Despite the low number of training data included for the neuronal network, the algorithm works very reliably. And thanks to the “taught” feed-forward network, it also correctly analyzes every test signal available. The benefits are twofold. Every situation in which an obstruction could occur is identified as such, and yet no situation during normal operation is incorrectly diagnosed as an obstruction. For every available test signal, the algorithm recognizes the lion’s share of obstructions between 200 and 500 ms (at the latest). Obstructions are also typified independent of the present rpm range and current seat position. Even better: different loads on the seat do not impact the algorithm’s analytic capacity.