SMART VEHICLE SEAT

Abstract

The invention concerns a system for controlling a smart vehicle seat (1) having a seat base (1b) and a backrest (1c), the surfaces of which are provided with flat pressure sensors, the pressure sensors being connected to a microcontroller (6) via which an actuator element is controllable via a drive unit (9), whereby the actuator element is adapted to change the surface of the seat base (1b) and/or the backrest (1c). It is the object of the invention to develop the system in a way that it automatically adapts to a sitting posture of a user and adjusts the surface of the seat base and of the backrest in an extremely flexible and exactly way depending on the user's posture. Therefore, the invention proposes, that the actuator element comprises one or more valves via which one or more airbags are controllable. Furthermore the invention proposes to use graphene sensors as pressure sensors and to integrate an artificial intelligence module (7) on the microcontroller (6).

Claims

1. System comprising a smart vehicle seat (1) having a seat base (1b) and a backrest (1c), the surfaces of which are provided with flat pressure sensors, the pressure sensors being connected to a microcontroller (6) via which an actuator element is controllable via a drive unit (9), whereby the actuator element is adapted to change the surface of the seat base (1b) and/or the backrest (1c), characterized in that the actuator element comprises one or more valves (9) via which one or more airbags (10) are controllable.

2. System according to claim 1, characterized in that the airbags (10) in the seat base (1b) and in the backrest (1c) are integrated below the surface.

3. System according to claim 1, characterized in that the pressure sensors are designed as graphene sensor arrays.

4. System according to claim 1, characterized in that the graphene sensor arrays consist of a plurality of sensor elements (3), wherein in each sensor element (3) graphene strip conductors (4) are arranged on a substrate (5).

5. System according to claim 4, characterized in that each single sensor element (3) is smaller than 0.5 mm.sup.2, preferably smaller than 0.37 mm.sup.2.

6. System according to claim 1, characterized in that an artificial intelligence module (7) is integrated in the microcontroller (6).

7. A method for controlling a system of a smart vehicle seat (1), in which the pressure distribution on the surface of the vehicle seat (1) is measured by means of pressure sensors, the measurement data are forwarded to a microcontroller (6), the microcontroller determines the sitting posture (P.sub.1, P.sub.2, P.sub.3, P.sub.4) on the basis of the measured data and learned knowledge using an artificial intelligence module (7) and controls via valves (9) an actuator element according the sitting posture (P.sub.1, P.sub.2, P.sub.3, P.sub.4).

8. The method according to claim 7, characterized in that the learned knowledge is based on a record which is processed by means of a recurrent neural network (RNN) and the sitting posture (P.sub.1, P.sub.2, P.sub.3, P.sub.4) is determined by means of the Softmax-function.

9. The method according to claim 8, characterized in that data input of the recurrent neural network (RNN) is encoded by a convolutional neural network (CNN).

10. The method according to claim 7, characterized in that the sitting posture (P.sub.1, P.sub.2, P.sub.3, P.sub.4) and/or the actuator settings are retrieved and/or set via a human-machine-interface (HMI).

11. The method according to claim 8, characterized in that the actuator element comprises airbags (10) integrated in the smart vehicle seat (1), wherein the valves (9) control the individual pressure in each airbag (10).

Description

[0022] The invention will be explained in the following in more detail with reference to the drawings.

[0023] FIG. 1a-d schematically show a smart vehicle seat of a system according to the invention in different load conditions;

[0024] FIG. 2 schematically shows the structure of a sensor array of a system according to the invention;

[0025] FIG. 3 schematically shows the structure of a sensor element of a system according to the invention;

[0026] FIG. 4 schematically shows a block diagram of a system according to the invention;

[0027] FIG. 5 schematically illustrates a first example of the determination of the sitting posture according to the method of the invention.

[0028] FIG. 6 schematically illustrates a second example of the determination of the sitting posture according to the method of the invention.

[0029] In FIGS. 1a-d, a vehicle seat is designated by reference numeral 1. The vehicle seat 1 has a seat base 1a and a backrest 1b and a headrest 1c. On the surface of the seat base 1a and the user-facing surface of the backrest 1b, a plurality of graphene sensors 2 are arranged. The graphene sensors 2 measure the pressure distribution caused by the user sitting on the vehicle seat 1.

[0030] The pressure distribution in FIGS. 1a-d varies with the sitting posture of the user. FIG. 1a shows the balanced pressure distribution (P1) in a straight and upright sitting position (sitting posture P1). In FIG. 1b, a higher pressure is measured on the right side because the user has shifted his weight to the right (sitting posture P2). A weight shift to the left is shown in FIG. 1c (sitting position P3). In FIG. 1d, the user obviously bends forward because only slight pressure is exerted on the backrest 1b (sitting posture P4). There can be determined four different sitting postures of the user in this embodiment. Practically, the number of sitting postures is of course not limited. In this way, the sitting behavior of the user can be monitored with the graphene sensors 2.

[0031] The sensors arranged on the vehicle seat 1 will be described in more detail as follows:

[0032] On the surfaces of the vehicle seat 1 flexible sensor arrays 2a are arranged with sensor elements 3 of graphene. Each sensor array 2a has a plurality of sensor elements 3 as shown in FIG. 2. The sensor elements 3 are printed on PET film 2b of 25 μm thickness. The size of the sensor array 2a is 300 mm×300 mm in this embodiment. There 100×100 sensor elements 3 arranged on the sensor array 2a. The sensor elements 3 are located 25 mm away from the edges at the four corners. Thus, it is possible generate 10000 pressure values.

[0033] Each individual sensor element 3 delivers its measured value normalized in the range of 0.0-1.0 for the respective measured pressure. For the realization, large-area sensor arrays of two-dimensional materials are used in this exemplary embodiment. According to the invention, graphene is used to obtain the desired pressure sensors. Thus, a flexible pressure sensor with a low operating voltage of less than 3.5 V, a high pressure sensitivity of 0-1 kPa and an excellent mechanical wear resistance over at least 3000 cycles is available.

[0034] FIG. 3 shows a single sensor element 3. A crosswise graphene strip conductor 4 is arranged on a substrate 5. The substrate 5 is, in this embodiment, a PET film 2b. If a force F acts on the substrate 5, this proportionally influences the electrical properties of the graphene strip conductor 4. Consequently, the magnitude of the force F can be determined. The single sensor element 3 is very small and flat. It extends only an area of 0.36 mm.sup.2 and a thickness of 25 μm. The graphene strip conductor 4 is 50 μm wide. On the graphene strip conductor 4 a voltage supply and an electrode for measuring the output voltage is arranged.

[0035] The properties of the PET film 2b as substrate used in this embodiment are as follows:

TABLE-US-00001 Density: 1420 kg/m.sup.3 Young's modulus: 2.5 GPa Poisson's ratio 0.34 Heat capacity: 1090 J/(kg*K) Thermal conductivity: 0.12 W//(m*K) Relative permittivity: 3.4  Resistivity: 1.5 * 10.sup.15 Ωm

[0036] The properties of Graphene used in this embodiment are as follows:

TABLE-US-00002 Density: 2200 kg/m.sup.3 Poisson's ratio: 0.16 Thermal conductivity: 5000 W/(m*K) Coefficient of thermal 8 * 10.sup.−6 1/K expansion: Relative permittivity: 2.14 Resistivity: 30 Ωm Electrical conductivity: 3.333 *10.sup.−2 S/m

[0037] FIG. 4 shows a block diagram which schematically illustrates the components of the system according to the invention and their interaction, whereby the measured pressure distribution is taken into account.

[0038] By means of the graphene sensors 2, the pressure distribution on the vehicle seat 1 is measured. The measured data are forwarded to a microcontroller 6. On the microcontroller 6, an artificial intelligence module 7 is integrated. The artificial intelligence module 7 determines the sitting posture on the basis of the measured data and on learned knowledge. The function of the artificial intelligence module 7 will be discussed in more detail below (see FIG. 4). The microcontroller 6 controls via a drive module 8 the valves 9, via which the integrated airbags 10 in the vehicle seat 1 can be pressurized. Which airbag 10 is subjected to how much pressure depends in each case on the sitting posture determined by the artificial intelligence module 7. The microcontroller 6 generates a PWM signal based on the determined sitting posture. Based on the PWM signal, the drive module 8 controls the valves 9.

[0039] Ultra-small solenoid air valves with extremely low weight are preferably used as valves 9 for low-pressure airbag controls. The airbags 10 have a very low weight, wherein the connections of the airbags 10 correspond to those of the valves 9. The outer fabric of the airbags 10 are glued to the seat foams of the vehicle seat 1.

[0040] The architectures according to the invention of the artificial neural networks which are preferably used in the artificial intelligence module 7 are shown schematically in FIGS. 5 and 6.

[0041] In an architecture according to the invention, shown in FIG. 5, a recurrent neural network (RNN) is used that can use the previous states (for example x.sub.t−1, x.sub.t−2, . . . , x.sub.tn) and computes the probability distribution of postures (P.sub.t) for a given set of data samples. For example, if the time series of size 5 is used, whereby it is possible to read the sensor data with 10 or 15 samples per second, then four data samples are used in the context (which is n) for predicting the posture of the current data sample x.sub.t. When using a RNN, the hidden units of the RNN are calculated differently than in a feedforward neural network (FNN), as RNNs have the ability to keep a memory from the previous states and use them to compute the current internal hidden state such as:

h.sub.t=σ(l.sub.h*x.sub.t+W.sub.h*h.sub.t−1+b.sub.h)

[0042] where, h.sub.t is the hidden internal state of the RNN; W.sub.h is a weight matrix, L.sub.h is an input projection matrix and b.sub.h is a bias vector, all these parameters are learned during training. Once, the internal hidden state is computed, the network use the softmax-function, to calculate the probability distribution of postures (P.sub.t):

P.sub.t=softmax (W.sub.p*h.sub.t+b.sub.p)

given

P∈(P.sub.1, P.sub.2, P.sub.3, P.sub.4) and Σ P=1

[0043] As the data is a time-series and sequential, the RNN method is supposed to provide better performance than a FNN. It is also productive to consider the form of the state values x.sub.t as matrices of an image. Then it is advantageous to use a convolutional neural network (CNNs) for encoding the state values and forward them to the RNNs, explained in FIG. 5.

[0044] The non-uniform pressure on the seat is expected to provide sparsity in the state values x.sub.t, that can make it difficult to feed the data as they are to the RNNs. Hence, a convolutional neural network (CNN) is used underneath the RNNs to encode the sensor signal data. As it can be seen that the grid sensor provides an image of the data, like frames at a 100×100 resolution. CNNs are most efficient to extract information from such an input stream of data. When using the CNN layer under the RNN layer, the architecture as shown in FIG. 6 can be configured. With this architecture, the CNN is used as an encoder to encode the input and then the encoded (in this case convoluted) input data are fed to the RNN which further uses that encoded data stream for final sitting behavior detection.

[0045] In both architectures (FIG. 5 and FIG. 6), the Softmax function calculates the probability distribution of possible current sitting postures (P.sub.1, P.sub.2, P.sub.3, P.sub.4) of the user. On the basis of the determined sitting postures (P.sub.1, P.sub.2, P.sub.3, P.sub.4) of the user of the vehicle seat 1, the artificial intelligence module 7 determines the optimum setting of the airbags 10, taking into account ergonomic aspects and a comfort analysis as well as the prediction of the user behavior. Based on the continuous measurement the artificial intelligence module 7 is continuously supplied with information in real time and thus trained further, so that the determination of the current sitting postures (P.sub.1, P.sub.2, P.sub.3, P.sub.4) becomes more and more reliable as a result.

[0046] Preferably, the system according to the invention has an interface to a higher-level server and to a user. The interfaces are essentially realized via the microcontroller 6.

[0047] Via communication with the higher-level-server, updates can be installed on the microcontroller 6. In addition, information can be shared. Thus, the microcontroller 6 can forward the measurement data regularly to the higher-level server. The higher-level server can then record the measured data from a large number of systems and develop behavioral rules and ergonomic procedures from the information obtained. With the help of these findings, the artificial intelligence module 7 can then be trained again on the microprocessor 6. The function of the artificial intelligence module 7 is improved even faster by providing and combining the data of a plurality of systems.

[0048] The user interface allows the user to provide feedback to the system. The user can also manually adjust the airbags 10 or select a preset program for controlling the airbags 10. Ergonomic aspects for supporting particularly back-friendly postures and variations of sitting postures can be taken into account in the given programs. The respective setting of the vehicle seat 1 can for example also be assigned to certain activities, such as “reading”, “driving” or “working”. Here, too, the respective settings may vary at intervals, taking into account ergonomic aspects. The feedback of the user can also be used to train the artificial intelligence module 7.

[0049] In summary, the invention teaches a novel combination of flexible high-sensitivity graphene-pressure-sensors and novel actuators with the use of artificial neural networks. The system according to the invention makes possible a comfort automation by means of a prediction and analysis of the user behavior. Such a system for a smart vehicle seat is in principle suitable for all seats used in any vehicles for the driver and for passengers. In particular, in the automotive and aviation industries, but also in trains and on ships, systems according to the invention are well suited to be applied for the control of an intelligent vehicle seat.

[0050] They are particularly attractive to people who spend a lot of time in vehicles and are therefore dependent on a certain level of comfort and on a support in their usual sitting behavior in order to prevent permanent physical problems, in particular back problems.

LIST OF REFERENCE NUMBERS

[0051] 1 vehicle seat

[0052] 1a seat base

[0053] 1b backrest

[0054] 1c headrest

[0055] 2 graphene sensors

[0056] 2a sensor array

[0057] 2b PET film

[0058] 3 sensor element

[0059] 4 graphene conductor strip

[0060] 5 substrate

[0061] 6 microcontrollers

[0062] 7 artificial intelligence module

[0063] 8 drive module

[0064] 9 valve

[0065] 10 airbag

[0066] P.sub.1-P.sub.4 sitting postures