DE102022208176A1 - Method for evaluating sensor data, computing unit for evaluating sensor data and sensor system, method for producing a sensor system - Google Patents

Abstract

The invention relates to a method for evaluating sensor data from at least one sensor. First, raw sensor data and/or processed sensor data and a process noise parameter of the sensor are read in. The process noise includes at least a time drift of a parameter of the sensor. The raw sensor data and/or the processed sensor data are then processed in a probabilistic filter, with the process noise being used to estimate a sensor error condition of the probabilistic filter. Estimated sensor errors are then removed from the raw sensor data and/or the processed sensor data and measurement data is determined from them. The measurement data can be calculated, for example, from the raw sensor data and/or the processed sensor data. Finally, the measurement data is output

Description

The invention relates to a method for evaluating sensor data, a computing unit set up to carry out the method, a sensor system and a method for producing a sensor system.

State of the art

Methods are known from the prior art in which raw sensor data or processed sensor data can be further processed using a probabilistic filter. It can be provided to correct the raw sensor data and/or the processed sensor data using a filter model and then to determine measurement data from the corrected raw sensor data or the corrected processed sensor data. The probabilistic filter can include, for example, a Kalman filter. In particular, such a method can be used to evaluate acceleration sensor data and rotation rate sensor data.

Disclosure of the invention

An object of the invention is to provide an improved method for evaluating sensor data, a computing unit for carrying out the method, an improved sensor system and a method for producing a sensor system. These tasks are solved with the subject matter of the independent patent claims. Advantageous further developments are specified in the dependent patent claims.

According to a first aspect, the invention comprises a method for evaluating sensor data of at least one sensor, which comprises the steps explained below. First, raw sensor data and/or processed sensor data are read in and a parameter of process noise of the sensor is read in. The process noise includes at least a time drift of a parameter of the sensor. The process noise parameter describes the time drift of the sensor's parameter. Raw sensor data can be unprocessed data from the sensor. If data processing is already planned within the sensor, processed sensor data is read in. Subsequently, the raw sensor data and/or the processed sensor data are processed in a probabilistic filter, wherein the process noise parameter is used to estimate a sensor error condition of the probabilistic filter. Estimated sensor errors are then removed from the raw sensor data and/or the processed sensor data and measurement data is determined from them. The measurement data can be calculated, for example, from the raw sensor data and/or the processed sensor data. Finally, the measurement data is output.

The fact that the sensor error state of the probabilistic filter is estimated using the process noise parameter results in a simple implementation of taking the process noise into account. If the process noise is different, for example for different sensors, a simple adaptation of the method is possible. The process noise or the parameters of the process noise can be determined experimentally and stored in a memory so that it is available during the execution of the method according to the invention.

According to a second aspect, the invention comprises a computing unit with an input, an output and a processor. The computing unit is set up to receive raw sensor data and/or processed sensor data via the input, then use the processor to carry out the method according to the invention for evaluating sensor data and then output the measurement data via the output.

According to a third aspect, the invention comprises a sensor system with a computing unit according to the invention and at least one sensor, wherein the sensor is set up to convert a physical measurement variable into raw sensor data and/or processed sensor data and to output it to the input of the computing unit.

According to a fourth aspect, the invention comprises a method for producing a sensor system, wherein the parameter of the process noise of the sensor is determined in a test environment with defined environmental parameters and then stored in a memory of the computing unit.

In one embodiment of the method for evaluating sensor data, the probabilistic filter comprises a Kalman filter. The Kalman filter can be designed, for example, as a nonlinear Kalman filter, i.e., among other things, as an extended Kalman filter or as a Cubature Kalman filter, in particular as a square-root Cubature Kalman filter. Alternatively, the probabilistic filter can be designed as an H-infinity filter or as a sequential Monte Carlo filter (SMC filter).

In one embodiment of the method for evaluating sensor data, the corrected sensor data is further integrated using the probabilistic filter to generate the measurement data If this is the case, the method according to the invention can be easily implemented.

In one embodiment of the method for evaluating sensor data, the process noise includes a time-defined shift of sensor values and a time-defined shift of a scale of the sensor. The temporally defined shift of sensor values can be referred to as temporal drift of the sensor values. The temporally defined shift in the scale of the sensor can be referred to as temporal scale drift.

The time-defined shift of sensor values and/or the time-defined shift of the scale of the sensor can have been determined for a single sensor. In this case, these values can be stored in memory for a single sensor. However, each sensor must then be evaluated in the test environment with defined environmental parameters during the process of producing a sensor system. Alternatively, it can be provided that a number of sensors, for example one hundred, are evaluated in the test environment with defined environmental parameters and then an average value is used for the time-defined shift of sensor values and/or the time-defined shift of the scale of the sensor.

In one embodiment of the method for evaluating sensor data, the process noise further comprises a shift in the sensor values depending on a measured variable and a shift in the scale depending on the measured variable. The measured variable can also have an influence on the accuracy of the sensor and thus influence the raw sensor data or the processed sensor data.

The shift of the sensor values depending on the measured variable and/or the shift of the scale depending on the measured variable can have been determined for a single sensor. In this case, these values can be stored in memory for a single sensor. However, each sensor must then be evaluated in the test environment with defined environmental parameters during the process of producing a sensor system. Alternatively, it can be provided that a number of sensors, for example one hundred, are evaluated in the test environment with defined environmental parameters and then an average value is used for the shift of the sensor values depending on the measured variable and/or the shift of the scale depending on the measured variable. In the test environment, the measurement variable can be set and changed, while other influencing factors are kept constant.

In one embodiment of the method for evaluating sensor data, the measured variable is a temperature. In the test environment, the temperature can be set and the shift of the sensor values depending on the temperature and/or the shift of the scale depending on the temperature can be determined for different temperatures.

In one embodiment of the method for evaluating sensor data, the process noise further comprises a further shift of the sensor values depending on a further measured variable and a further shift of the scale depending on the further measured variable. This can be implemented analogously to the measurement variable.

In one embodiment of the method for evaluating sensor data, the further measurement variable is air humidity. In the test environment, the humidity can be set and the shift of the sensor values depending on the humidity and/or the shift of the scale depending on the humidity can be determined for different temperatures.

In addition to temperature and humidity, other variables are also conceivable as measured variables and/or further measured variables. These can include, for example, a pressure, a thermal voltage, or a time gradient of temperature, humidity, pressure and/or thermal voltage.

In one embodiment of the method for evaluating sensor data, the sensor is a rotation rate sensor. Furthermore, raw sensor data and/or processed sensor data from an acceleration sensor are read in. Using the probabilistic filter, a position and a speed are determined as measurement data. The method according to the invention can be used both for the rotation rate sensor and for the acceleration sensor. In addition, a magnetic field sensor can also be provided, which is also evaluated using the method according to the invention. In particular, the time-defined shift of sensor values and/or the time-defined shift of the scale of the sensor can be determined for the acceleration sensor, optionally expanded to include the shift of the sensor values depending on the measured variable and/or the shift of the scale depending on the measured variable and, if necessary, the Shift of the sensor values depending on the other measured variable and/or the shift of the scale depending on the other measured variable. The The measured variable can be in particular the temperature, the further measured variable in particular the humidity. The same applies to the acceleration sensor and, if available, also to the magnetic field sensor. The rotation rate sensor can be designed or referred to as a gyroscope.

In one embodiment of the sensor system, the sensor is a rotation rate sensor. The sensor system further includes an acceleration sensor. The acceleration sensor is set up to convert a physical measurement variable into raw sensor data and/or processed sensor data and output it to the input of the computing unit. In addition, it can be provided that the sensor system also includes a magnetic field sensor.

• 1 ein Ablaufdiagramm eines Verfahrens zum Auswerten von Sensordaten;
• 2 ein Sensorsystem;
• 3 eine Testumgebung, die während eines Verfahrens zum Herstellen eines Sensorsystems genutzt werden kann; und
• 4 eine Übersicht über ein Sensorsystem.

Embodiments of the invention are explained using the following drawings. Show in the schematic drawing:

• 1 a flowchart of a method for evaluating sensor data;
• 2 a sensor system;
• 3 a test environment that may be used during a process for manufacturing a sensor system; and
• 4 an overview of a sensor system.

1 shows a flowchart 100 of a method for evaluating sensor data of at least one sensor. The following steps are carried out. First, raw sensor data and/or processed sensor data are read in in a first method step 101. The raw sensor data can be the analog or digital data output by the sensor. The processed sensor data can be further processed analog or digital data. Furthermore, raw sensor data and/or processed sensor data from more than one sensor can also be read in in the first method step 101. In a second method step 102, a parameter of a process noise of the sensor is read in, the process noise comprising at least a time drift of a parameter of the sensor. If raw sensor data and/or processed sensor data from more than one sensor have been read in in the first method step 101, the process noise parameter can also be read in for more than one sensor, in particular for all sensors, in the second method step 102. Subsequently, the raw sensor data and/or the processed sensor data are processed in a probabilistic filter in a third method step 103, wherein the parameter of the process noise is used to estimate a sensor error state of the probabilistic filter. The probabilistic filter can in particular have states for the raw sensor data or the processing sensor data. In a fourth method step 104, estimated sensor errors are removed from the raw sensor data and/or the processed sensor data and measurement data is determined therefrom. The measurement data can, on the one hand, contain corrected sensor data, but on the other hand also derived values determined from the sensor data. For example, a speed and/or a position could be determined as derived values from the sensor data of an acceleration sensor. In a fifth method step 105, the measurement data are output.

The fact that the sensor error state of the probabilistic filter is estimated using the process noise parameter results in a simple implementation of taking the process noise into account. If the process noise is different, for example for different sensors, a simple adaptation of the method is possible. The parameter of the process noise can be determined experimentally and stored in a memory so that it is available during the execution of the method according to the invention.

In an exemplary embodiment of the method for evaluating sensor data, the probabilistic filter used in the third method step 103 comprises a Kalman filter. The Kalman filter can be designed, for example, as a nonlinear Kalman filter, i.e., among other things, as an extended Kalman filter or as a Cubature Kalman filter, in particular as a square-root Cubature Kalman filter. Alternatively, the probabilistic filter can be designed as an H-infinity filter or as a sequential Monte Carlo filter (SMC filter).

In an exemplary embodiment of the method for evaluating sensor data, the corrected sensor data are further integrated using the probabilistic filter in the fourth method step 104 in order to generate the measurement data. This is particularly easy to implement for the Kalman filter, with both the third method step 103 and the fourth method step 104 then being carried out using the Kalman filter.

In one embodiment of the method for evaluating sensor data, the process noise includes a time-defined shift of sensor values and a time-defined shift of a scale of the sensor. The temporally defined shift of sensor values can be referred to as temporal drift of the sensor values. The time-defined shift in the scale of the sensor can be called time scale drift be referred to. For these types of process noise, a process noise parameter can be determined and used in the process.

In one embodiment of the method for evaluating sensor data, the process noise further includes a shift in the sensor values depending on a measured variable and a shift in the scale depending on the measured variable. The measured variable can also have an influence on the accuracy of the sensor and thus influence the raw sensor data or the processed sensor data. For these types of process noise, a process noise parameter can be determined and used in the process.

In one embodiment of the method for evaluating sensor data, the measured variable is a temperature. This means that a shift in the sensor values that occurs with most sensors depending on the temperature and a shift in the scale that occurs with most sensors depending on the temperature can be taken into account and thus more precise measurement data can be achieved.

In one embodiment of the method for evaluating sensor data, the process noise further comprises a further shift of the sensor values depending on a further measured variable and a further shift of the scale depending on the further measured variable. In this way, other measurement variables that may have an influence can be taken into account. For these types of process noise, a process noise parameter can be determined and used in the process.

In an exemplary embodiment of the method for evaluating sensor data, the further measurement variable is humidity. This means that a shift in the sensor values that occurs with most sensors depending on the humidity and a shift in the scale that occurs with most sensors depending on the humidity can be taken into account and thus more precise measurement data can be achieved.

In one embodiment of the method for evaluating sensor data, the sensor is a rotation rate sensor. Furthermore, raw sensor data and/or processed sensor data from an acceleration sensor are read in in the first method step 101. In the second method step 102, the process noise of the rotation rate sensor and the process noise of the acceleration sensor are then read in. Using the probabilistic filter, a position and a speed are determined as measurement data in the fourth method step 104.

2 shows a sensor system 110, which has a rotation rate sensor 111, an acceleration sensor 112 and a magnetic field sensor 113. It can be provided that only one of the sensors 111, 112, 113 mentioned is present. Furthermore, one or more alternative sensors, for example a GPS sensor, can also be provided. The sensors 111, 112, 113 are set up to convert a physical measurement variable (a rotation rate for the rotation rate sensor 111, an acceleration for the acceleration sensor 112 and a magnetic field for the magnetic field sensor 113) into raw sensor data and/or processed sensor data and to inputs 121, 122, 123 of the computing unit to output. The rotation rate sensor 111 is connected to a first input 121 of a computing unit 120. The acceleration sensor 112 is connected to a second input 122 of the computing unit 120. The magnetic field sensor 113 is connected to a third input 123 of the computing unit 120. The number of inputs 121, 122, 123 can be adapted to the number of sensors 111, 112, 113. The computing unit 120 also has a processor 124. The computing unit is set up to receive raw sensor data and/or processed sensor data via the inputs 121, 122, 123, then in connection with this by means of the processor 124 1 carry out the procedures explained and then output the measurement data via an output 125.

Furthermore, in 2 shown that the computing unit 120 also optionally has a fourth input 126 and a memory 127. It can be provided that the parameter of the process noise of the sensors 111, 112, 113 is determined in a test environment with defined environmental parameters, provided via the input 126 and then stored in the memory 127.

$$Q_{gyroScale}=dt\cdot S{D}_{gyro}+{\Delta }_{Temp}\cdot TC{S}_{gyro}$$

$$Q_{accBias}=dt\cdot O{D}_{acc}+{\Delta }_{Temp}\cdot TC{O}_{acc}$$

$$Q_{accScale}=dt\cdot S{D}_{acc}+{\Delta }_{Temp}\cdot TC{S}_{acc}$$

$$Q_{magBias}=dt\cdot O{D}_{mag}+{\Delta }_{Temp}\cdot TC{O}_{mag}$$

$$Q_{magScale}=dt\cdot S{D}_{mag}+{\Delta }_{Temp}\cdot TC{S}_{mag}$$

For the in 2 Sensor system 110 shown can, for example, be provided that a Kalman filter is used as a probabilistic filter during the method carried out by means of processor 124. If this is the case, the process noise can be taken into account in the form of a noise model with Q values for all relevant variables. The noise model can be referred to as a covariance matrix Q, whereby individual components of the covariance matrix can be determined using the following equations: $$Q_{GyrObias}=dt\cdot O{D}_{GyrO}+{\Delta }_{Temp}\cdot TC{O}_{GyrO}$$

$$Q_{GyrOScale}=dt\cdot S{D}_{GyrO}+{\Delta }_{Temp}\cdot TC{S}_{GyrO}$$

$$Q_{accbias}=dt\cdot O{D}_{acc}+{\Delta }_{Temp}\cdot TC{O}_{acc}$$

$$Q_{accScale}=dt\cdot S{D}_{acc}+{\Delta }_{Temp}\cdot TC{S}_{acc}$$

$$Q_{maGbias}=dt\cdot O{D}_{maG}+{\Delta }_{Temp}\cdot TC{O}_{maG}$$

$$Q_{maGScale}=dt\cdot S{D}_{maG}+{\Delta }_{Temp}\cdot TC{S}_{maG}$$

The subscript “gyro” refers to the rotation rate sensor 111. The subscript “acc” refers to the acceleration sensor 112. The subscript “mag” refers to the magnetic field sensor 113. The subscript “Bias” refers to the sensor values as such. The subscript “Scale” refers to a scale of sensor values. OD corresponds to a time-defined shift in the respective sensor values (offset drift). SD refers to a time-defined shift in a scale of the sensor (scale drift). TCO is a change in the shift of the respective sensor values depending on the temperature (offset change over temperature). TCS is a change in the shift of the sensor's scale (scale change over temperature). OS, SD, TCO and TCS can be experimentally determined values for a sensor 111, 112, 113 or can be determined by averaging over several sensors 111, 112, 113. Using dt, a time difference between two runs of the Kalman filter is expressed. A temperature difference is taken into account using Δ _Temp . If only a temporal component is intended to be taken into account in the covariance matrix, the temperature-dependent terms can be omitted. Furthermore, instead of temperature as a measured variable, another measured variable with analogous formulas can also be taken into account. In addition, it is possible to take another measurement variable into account in addition to temperature as a measurement variable by integrating additional corresponding terms into the formulas.

3 shows a test environment 130 in which six rotation rate sensors 111 are arranged, although more or fewer than six rotation rate sensors 111 can also be arranged. This test environment 130 can be used for a method of producing a sensor system 110, for example as shown in 2 shown can be used. The parameter of the process noise of the rotation rate sensors 111 is determined in the test environment 130 with defined environmental parameters and then stored in the memory 127 of the computing unit 120. Instead of the rotation rate sensors 111, several acceleration sensors 112 or several magnetic field sensors 113 can also be arranged in the test environment 130.

The time-defined shift of sensor values and/or the time-defined shift of the scale of the sensor can be determined for an individual rotation rate sensor 111. In this case, these values can be stored in memory 127 for an individual rotation rate sensor 111, so that each sensor system 110 receives an individual value. However, each rotation rate sensor 111 must then be evaluated in the test environment 130 with defined environmental parameters in the method for producing a sensor system 120. Alternatively, it can be provided that a number of rotation rate sensors 111, for example one hundred, are evaluated in the test environment 130 with defined environmental parameters and then an average value is used for the time-defined shift of sensor values and/or the time-defined shift of the scale of the sensor. The same procedure can be followed for the acceleration sensors 112 or the magnetic field sensors 113.

Analogously, the shift of the sensor values depending on the measured variable and/or the shift of the scale depending on the measured variable can have been determined for an individual rotation rate sensor 111. In this case, these values can be stored in memory 127 for a single yaw rate sensor 111. However, each yaw rate sensor 111 must then be evaluated in the test environment 130 with defined environmental parameters during the method for producing a sensor system 120. Alternatively, it can be provided that a number of rotation rate sensors 111, for example one hundred, are evaluated in the test environment 130 with defined environmental parameters and then an average value is used for the shift of the sensor values depending on the measured variable and/or the shift of the scale depending on the measured variable . In the test environment 130, the measurement variable can be set and changed, with other influencing factors being kept constant. The method according to the invention can be used both for the rotation rate sensor 111 and for the acceleration sensor 112. In addition, a magnetic field sensor 113 can also be provided, which is also evaluated using the method according to the invention. In particular, the time-defined shift of sensor values and/or the time-defined shift of the scale of the sensor can be determined for the rotation rate sensor 111, the acceleration sensor 112 and the magnetic field sensor 113, optionally expanded to include the shift of the sensor values depending on the measured variable and/or the shift the scale depending on the measured variable and, if necessary, the shift of the sensor values depending on the further measured variable and/or the shift of the scale depending on the further measured variable. The measured variable can be in particular the temperature, the further measured variable in particular the humidity.

4 shows an overview 135 of a sensor system 120, which corresponds to the sensor system 120 of 2 can correspond. First there is a sensor characterization 140 carried out, for example in connection with 3 explained. By means of further sensors, here for example a temperature sensor 114, a humidity sensor 115, a pressure sensor 116 and a stress sensor 117, information about the characterized behavior of the rotation rate sensor 111, the acceleration sensor 112 and the magnetic field sensor 113 can be obtained. The stress sensor 117 can be set up to measure mechanical stress or thermal stress. Not all of them necessarily have to be in 4 additional sensors 114, 115, 116, 117 shown can be provided. In a covariance matrix determination 141, the parameter of the process noise 142 is then determined from the sensor characterization 140 and the data from the further sensors 114, 115, 116, 117, for example by means of the in connection with 2 explained equations. In a probabilistic filter 143, for example a Kalman filter, the parameter of the process noise 142 is further processed, together with the raw sensor data and / or processed sensor data of the yaw rate sensor 111, the acceleration sensor 112 and the magnetic field sensor 113. Position, speed and Orientation of the sensor system 120 are output. The other sensors 114, 115, 116, 117 can also be in the sensor system 120 2 be arranged, if necessary individually in such a way that only one temperature sensor 114 is arranged.

Although the invention has been described in detail by the preferred embodiments, the invention is not limited to the examples disclosed and other variations may be derived therefrom by those skilled in the art without departing from the scope of the invention.

Claims (10)

Method for evaluating sensor data of at least one sensor (111, 112, 113), with the following steps: - Reading raw sensor data and/or processed sensor data; - reading in a parameter of a process noise (142) of the sensor (111, 112, 113), the process noise (142) comprising at least one time drift of a parameter of the sensor (111, 112, 113); - Processing the raw sensor data and/or the processed sensor data in a probabilistic filter (143), wherein the process noise parameter (142) is used to estimate a sensor error state of the probabilistic filter (143); - Removing estimated sensor errors from the raw sensor data and/or the processed sensor data and determining measurement data therefrom; - Outputting the measurement data. Procedure according to Claim 1 , wherein the corrected sensor data is further integrated using the probabilistic filter (143) to generate the measurement data. Procedure according to one of the Claims 1 or 2 , wherein the process noise (142) comprises a time-defined shift of sensor values and a time-defined shift of a scale of the sensor (111, 112, 113). Procedure according to Claim 3 , wherein the process noise (142) further comprises a shift in the sensor values depending on a measured variable and a shift in the scale depending on the measured variable. Procedure according to Claim 4 , where the measured variable is a temperature. Procedure according to Claim 4 or 5 , wherein the process noise (142) further comprises a further shift of the sensor values depending on a further measured variable and a further shift of the scale depending on the further measured variable. Procedure according to one of the Claims 1 until 6 , wherein the sensor is a rotation rate sensor (111) and raw sensor data and/or processed sensor data from an acceleration sensor (112) are also read in, and a position and a speed are determined as measurement data by means of the probabilistic filter (143). Computing unit (120) with an input (121, 122, 123), an output (125) and a processor (124), the arithmetic unit (120) being set up to receive raw sensor data and/or via the input (121, 122, 123). to receive processed sensor data, then use the processor (124) to carry out the method according to one of the Claims 1 until 7 and then output the measurement data via the output (125). Sensor system (110) with a computing unit (120). Claim 8 and at least one sensor (111, 112, 113), wherein the sensor (111, 112, 113) is set up to convert a physical measurement variable into raw sensor data and/or processed sensor data and to output it to the input (121, 122, 123) of the computing unit . Method for producing a sensor system (110). Claim 9 , wherein the parameter of the process noise of the sensor (111, 112, 113) in a test environment (130) with a defined environment Exercise parameters are determined and then stored in a memory (127) of the computing unit (120).

Images