Sensor device for anomaly detection and procedures

Abstract

A sensor device (10) for anomaly detection is provided, comprising a sensor element (20) for detecting objects (T1, Tn) or for detecting a medium, and an evaluation unit (AE) for generating corresponding measured values ​​of a measurand, in particular position, distance, or process values, means for acquiring these measured values, and at least one of the following quantities: i. the first derivative with respect to the time of the measured value; ii. the second derivative with respect to the time of the measured value; iii. the integral over the time of the measured value. The sensor device further comprises means for parallel evaluation of the acquired quantity, means for determining switching information for each evaluated quantity, and means for logically combining the switching information for anomaly detection. A method for anomaly detection is also specified.

Description

[0001] The invention relates to a sensor device for anomaly detection according to claim 1 and a method for anomaly detection according to claim 4.

[0002] Various types of sensor devices are used in industrial process monitoring to collect data. Common sensor types include position sensors, distance sensors, and process sensors, each fulfilling specific tasks.

[0003] Position sensors are used, for example, to determine the position of machine components or workpieces. Distance sensors measure distances and help to avoid collisions or enable precise positioning. Process sensors monitor parameters such as the fill level, flow rate, and / or temperature of a medium, thus ensuring compliance with process specifications.

[0004] One example of the use of position sensors is monitoring the position of goods on conveyor belts in logistics. Distance sensors can be used in robotics to detect and avoid obstacles. Process sensors are used, for example, in the food industry to ensure the quality and safety of products.

[0005] Inductive sensors detect the approach of metallic objects without physical contact using a sensing element. Capacitive sensors can detect the approach of non-metallic objects and liquids using a sensing element. Magnetic sensors monitor and measure the position and movement of magnetic fields using a sensing element. Optical sensors use light in the visible and infrared ranges to detect objects and measure distances using a sensing element. They utilize light sources such as LEDs or lasers to direct light beams at a target object. The reflected light is then captured by a detector that analyzes the intensity and angle of the returning light. Radar and ultrasonic sensors measure objects, distances, and their movements using reflected echoes from the objects to a sensing element.

[0006] The analysis of continuously recorded measurements from all these sensor types can provide information about the position, distance and speed of objects, or information about the temporal progression of the measurements.

[0007] For anomaly detection, i.e., the detection of anomalies or malfunctions within an application or environment, the data generated by the sensors is collected and subjected to data analysis. Typically, this data analysis takes place outside the individual sensor device. For this purpose, measurement or raw data is transmitted to a higher-level entity located physically separate from the sensor device, such as a higher-level controller or a cloud system. By aggregating and analyzing a large amount of data, a more accurate assessment of what constitutes normal behavior and what constitutes a malfunction is achieved. A disadvantage of this approach is the large volume of data that must be transmitted. This transmission introduces a time delay, which limits real-time capabilities.

[0008] The key advantage of decentralized data analysis, performed spatially outside the sensor device, lies in its ability to integrate and analyze large volumes of data from diverse sources. This data aggregation allows for a more detailed and comprehensive understanding of normal operating conditions. For example, data from various sensors within a production process can be combined to gain a holistic understanding of system behavior. This enables the identification of patterns and trends that might otherwise remain hidden from isolated data sources.

[0009] Despite these advantages, decentralized data analysis also has significant drawbacks. Firstly, there is the risk of data corruption through aggregation. Furthermore, combining data can lead to the loss of important details or dilution of the overall dataset. This is particularly problematic when trying to identify anomalies, rare events, or malfunctions. For example, a minor but critical deviation in sensor data could be overlooked when aggregated into a large dataset. This corruption can result in anomalies either going undetected or being misinterpreted, compromising the reliability and accuracy of the analysis.

[0010] Another problem is the latency caused by the transmission of large amounts of data. The data must be transferred from the sensors to a higher-level database, which takes time. In many applications, especially in safety-critical areas, however, rapid anomaly detection is crucial. Delays in data transmission and analysis can lead to anomalies not being detected in time, which can have serious consequences.

[0011] In current technology, neural networks are used to analyze sensor data and recognize complex patterns and their relationships, which is often difficult with traditional methods. The use of neural networks is particularly advantageous in dynamic and unpredictable environments, such as those found in position and distance tracking, the process industry, mobile applications, or robotics.

[0012] For effective recognition of complex situations, AI systems must not only identify individual objects or events, but also understand the relationships and connections between them. The accuracy of these systems depends heavily on the quality of the training data. Incomplete or faulty data can quickly lead to incorrect results. A disadvantage is that data processing typically takes place outside the sensor device housing, which prevents real-time processing.

[0013] An anomaly can be detected by a single sensor element in a sensor device. Therefore, aggregating data from multiple sensor devices in an environment to detect a specific anomaly in an ecosystem is often counterproductive, as the data aggregation makes accurate anomaly detection more difficult.

[0014] The object of the invention is to avoid the aforementioned disadvantages of the prior art and to provide a sensor device and a method for detecting anomalies, which enables reliable and fast detection of anomalies.

[0015] The problem is solved with respect to the sensor according to claim 1 and with respect to the method according to claim 4. Advantageous embodiments of the invention are specified in the dependent claims.

[0016] The method presented here makes it possible to monitor a multitude of complex situations using data analysis and to detect anomalies—that is, malfunctions that differ from normal behavior—distinguish between them, and output them as switching information from a sensor device. The data analysis generates a customized parameter set. Because the data analysis functionality is implemented directly in the sensor device and not in the higher-level controller, real-time detection is enabled. In this context, real-time means that detection occurs within the sensor device's measurement interval. If the data were conventionally transmitted to the higher-level controller via IO-Link and evaluated there, this real-time capability would be lost.

[0017] The sensor device acquires the measured quantity, its first derivative, its second derivative, and its integral, and evaluates these measurements in parallel. The acquisition data for the neural network, or for the heuristic, can thus be continuously kept up-to-date and updated.

[0018] Furthermore, integrating data analytics into the sensor device offers several advantages. Integration reduces latency because data transmission to the higher-level control system occurs in real time. It increases system reliability because fewer components are involved, thus minimizing potential sources of error. Local data processing relieves the decentralized, higher-level control unit, leading to more efficient resource utilization. Real-time analysis enables an immediate response to detected anomalies, which is particularly beneficial in safety-critical applications.

[0019] The sensor device according to the invention for anomaly detection comprises a sensor element for detecting objects or for detecting a medium, an evaluation unit for generating corresponding measured values ​​of a measurand, in particular position, distance, or process values, and means for acquiring these measured values ​​and at least one of the following quantities: i. the first derivative with respect to the time of the measured value; ii. the second derivative with respect to the time of the measured value; iii. the integral over the time of the measured value. Furthermore, the sensor element comprises means for parallel or simultaneous evaluation of the acquired quantities, means for determining switching information for each evaluated quantity, and means for logically combining the switching information for anomaly detection.

[0020] These measured values, derived from the measured value, are evaluated together with the measured value, and a switching signal is determined for each value. The switching signals are logically combined to determine whether a fault, i.e., an anomaly, is present. The logical combination is parameterizable, for example, using Boolean operators. This allows, for instance, the input to be switched on or off.

[0021] The advantage of parallel data processing in the sensor device, starting with the measured value, the first and second derivatives, and the integral over time, is that it provides more measurement data. This is because you not only have the measured value, i.e., the measured quantity, but also information about the change in that quantity. This allows for the identification of more complex phenomena than previously possible.

[0022] By evaluating the integral, a distance sensor can be converted into a quantity sensor. Alternatively, a flow sensor could be used to create a volume sensor. In this case, the flow sensor could measure not only the volume, but also the fluctuations and velocity of the flow.

[0023] It is advantageous if the sensor device has multiple channels to detect different conditions simultaneously. This enables more comprehensive monitoring and analysis of complex situations, as multiple parameters can be acquired and evaluated in parallel. This increases the accuracy and reliability of anomaly detection. Furthermore, the sensor device can be used more flexibly because it is capable of monitoring different scenarios at the same time.

[0024] It is equally advantageous if the means for logically linking the switching information include a parameterizable logic unit. Here, the parameters for the evaluations and the type of logical connections are determined through data analysis of the acquired measured values ​​and derived quantities. This leads to dynamic adaptability of the system, as it is able to adjust automatically to changing conditions. Furthermore, the parameterizable logic unit allows for individual configuration tailored to specific requirements and applications. Additionally, the integration of such a logic unit reduces the need for external control units, which in turn lowers the complexity and cost of the overall system. A parameterizable logic unit (PLU) is understood to be a digital circuit that can be adapted by parameters.It allows the configuration of logical functions without physically changing the hardware.

[0025] It is advantageous if the sensor device incorporates a neural network as an evaluation unit, to which the acquired and derived values ​​are fed in order to recognize complex situations. A neural network, through its pattern recognition and learning capabilities, can perform a more precise and robust analysis of the acquired data. This allows even subtle anomalies and complex relationships, which are difficult to detect using conventional methods, to be reliably identified.

[0026] Additionally, using a neural network offers the advantage that the system can be continuously improved. Through regular training with new data, the neural network can further increase its accuracy and efficiency over time. This leads to a steady improvement in recognition performance and adaptability to new or changing conditions.

[0027] Another advantage is the reduction of the need for manual calibration and adjustment. The neural network can learn and adapt automatically. Furthermore, implementing a neural network in the sensor device enables centralized data processing, which minimizes latency and improves the system's real-time capability.

[0028] Finally, a neural network, due to its ability to process large amounts of data, can simultaneously analyze multiple parameters and their interactions. This expands the sensor device's application possibilities and makes it more versatile and powerful in monitoring and analyzing complex situations.

[0029] Furthermore, a method for detecting anomalies with such a sensor device is provided, comprising the steps of: acquiring a measured value by at least one sensor element integrated into the sensor device with an evaluation circuit (AE); acquiring at least one of the following quantities in parallel or simultaneously by the sensor device: i. the first derivative with respect to the time of the measured value, ii. the second derivative with respect to the time of the measured value, iii. the integral over the time of the measured value; evaluating the acquired quantities in parallel within the sensor device; determining a switching signal for each evaluated quantity; combining the switching signals using parameterizable operators to determine the presence of an anomaly.

[0030] The parallel or simultaneous acquisition of various derived quantities enables a more comprehensive data analysis, leading to more precise and robust anomaly detection. This multidimensionality of data analysis increases the reliability and accuracy of the system. The parallel evaluation of the acquired quantities within the sensor device reduces latency and enables real-time analysis. This is particularly advantageous in time-critical applications, as it allows for rapid responses to detected anomalies.

[0031] Determining switching information for each evaluated parameter allows for a differentiated and detailed assessment of the acquired data. This leads to a better distinction between normal and anomalous states and increases the precision of anomaly detection. Linking the switching information using parameterizable operators offers high flexibility and adaptability of the system. The operators can be individually configured to meet specific requirements and conditions.

[0032] It is advantageous to combine switching information using Boolean operators or binary logic. This method enables simple and efficient logic that is easy to implement. By using Boolean operators, the system can make clear and unambiguous decisions. Furthermore, computing power is used efficiently, as Boolean operations require fewer resources than more complex mathematical calculations. In addition, binary logic allows for simple and fast processing of switching information, which supports the system's real-time capability.

[0033] It is advantageous if the sensor device outputs the measured values ​​and derived quantities in addition to the switching information. This provides a more comprehensive data basis for further analysis and diagnostics. By providing the raw data and derived quantities, more in-depth investigations can be carried out and the causes of anomalies can be better understood. This increases the traceability of anomaly detection.

[0034] Furthermore, the output of measured values ​​and derived parameters enables easy integration into higher-level systems and databases. This facilitates centralized data storage and processing, as well as the use of the data for long-term trend analyses and optimization measures. Finally, providing this additional information contributes to improved maintenance and service processes, as accurate and detailed data is available to identify and resolve sources of error more quickly.

[0035] It is advantageous if the parameters for the evaluations and the type of logical connections are determined by data analysis of the acquired measurements, the derived quantities, or the integral over the time of the measured value. This method enables dynamic adaptation of the system to different operating conditions and applications.

[0036] Continuous analysis of the collected data allows for the automatic optimization of parameters and logical connections, thereby increasing the accuracy of anomaly detection. Because the parameters and logical connections are determined based on actual measurement data, the system can be adapted to a wide variety of scenarios and environments. This enables the sensor device to be used in diverse applications, from industrial process monitoring to environmental monitoring.

[0037] In addition, data-driven determination of evaluation parameters reduces the need for manual calibration and adjustment. This saves time and resources, as the system can adapt automatically to changing conditions. This also minimizes the risk of human error during configuration, increasing the system's reliability and security.

[0038] Through continuous data analysis and parameter adjustments, the system can steadily improve its performance over time. This leads to long-term optimization of anomaly detection and increases the efficiency of the sensor device.

[0039] It is advantageous for the sensor device to have multiple channels in order to simultaneously detect and output different states. By capturing and evaluating various conditions in parallel, the system can better understand complex situations and react more precisely to different scenarios.

[0040] It is advantageous if the derived quantities and the integral are additionally fed to a neural network implemented in the evaluation unit of the sensor device, whereby the data analysis takes place in the sensor device, so that the raw data does not have to be transferred to a higher-level control unit.

[0041] This offers several advantages. Firstly, the system's real-time capability is improved because data analysis takes place directly in the sensor device, eliminating delays caused by data transmission. This is particularly beneficial for time-critical applications. Furthermore, local data processing relieves the load on the higher-level control unit, leading to more efficient resource utilization and improved overall system performance.

[0042] Neural networks are capable of recognizing complex patterns and relationships in data, further increasing the accuracy and reliability of anomaly detection. Through continuous adaptation and improvement of the neural network, the system's performance can be further optimized over time. The measured signals, the generated derivatives, and the additionally acquired parameters are used directly for training and evaluation.

[0043] The invention will now be explained in more detail using exemplary embodiments and with reference to the drawings.

[0044] In the following description of preferred embodiments, identical reference numerals denote identical or comparable components.

[0045] They show schematically: Fig. 1 a schematic representation of a sensor arrangement of a sensor device; Fig. 2 a schematic representation of a further sensor arrangement according to Fig. 1 in a time course; Fig. 3 a diagram showing the output of the measured quantities; Fig. 4. A method for detecting anomalies; Fig. 5. Another method for detecting anomalies.

[0046] The in Fig. The sensor arrangement and time sequence shown in Figure 1 are exemplary for a detection task of a sensor device 10, for example, a position, distance, or process sensor device, which detects an object T1, Tn and generates a measured value using an inductive, capacitive, magnetic, optical principle, or by detecting reflected echoes. The sensor arrangement shows a sensor device 10 that detects several objects, so-called targets T1, Tn. The sensor device 10 has a single sensor element 20 and an associated evaluation unit AE for generating position or distance values. As shown, the targets T1, Tn differ in size or thickness and are located at different distances from the stationary sensor device 10.By moving targets T1 and Tn, which are spaced at a fixed distance from each other, along a path to the right or left past the sensor device 10, the sensor element 20, located within the sensor device 10, individually detects each target. The respective measured value P of target T1 and target Tn is shown in a diagram below the array. A window is defined in the diagram for each switching output S1 and Sn. Each window covers a range of the measured value P. Normally, if a measured value P falls within a window, the switching output would activate, as shown in the diagram. This usually results in a false trigger. To prevent this false trigger, a filter F is placed upstream of the sensor device 10. In filter mode F, the false trigger is suppressed, allowing the additional target Tn to be detected and evaluated alongside target T1.According to the invention, this prevents incorrect switching and sets the outputs S1 and Sn to a target-specific measured value P. If filter mode F is active, the switching outputs S1 and Sn are only switched if a time-constant measured value P is detected within the respective window of the switching output S1 or Sn. For this purpose, the sensor device comprises 10 means for acquiring a measured value and at least one of the following quantities: the first derivative with respect to the time of the measured value, the second derivative with respect to the time of the measured value, or the integral over the time of the measured value. Furthermore, the sensor device comprises 10 means for parallel and simultaneous evaluation of the acquired quantity, means for determining switching information for each evaluated quantity, and means for logically combining the switching information to detect malfunctions, i.e., an anomaly, and to automatically adjust the parameters of filter F.

[0047] Fig. Figure 2 shows an example of another sensor arrangement. The sensor device 10, for example a position, distance or process sensor device, which detects an object by means of an inductive, capacitive, magnetic, optical principle or by means of detecting reflected echoes, exhibits, as in Fig. 1 a single sensor element 20 and an associated evaluation unit AE. In Fig. In Figure 2, targets T1 and Tn are arranged at a predefined distance from each other, and the sensor device 10 moves along a defined path past targets T1 and Tn. This principle also works if the sensor device 10 is stationary and the targets are moved past the sensor device 10. If filter mode F is active, the switching outputs S1 and Sn only activate if a constant measured value P is detected within the respective window of the switching output S1 and Sn. If the measured value P changes too much over time, the switching outputs S1 and Sn are not activated. The filter F prevents false activation. For this purpose, the sensor device 10 includes, as described in Figure 2, the following components: Fig. 1. Means for acquiring a measured value and at least one of the following quantities: the first derivative with respect to the time of the measured value, the second derivative with respect to the time of the measured value, the integral over the time of the measured value. In addition, the sensor device includes 10 means for parallel and simultaneous evaluation of the acquired quantity, means for determining switching information for each evaluated quantity, and means for logically combining the switching information to detect a malfunction, i.e., an anomaly, and to automatically adjust the parameters of the filter F.

[0048] Fig. Figure 3 illustrates the possible output of the determined measured quantities at the respective switching outputs S1 and Sn. The output can be in the form of measured quantities or as binary switching information. Binary switching information is output after evaluation in the evaluation unit AE. A prior derivation in the evaluation unit enables the output of the binary switching information of this derivation, or the value of the time derivative of this measured quantity. The output of the measured quantity with the corresponding switching information can occur simultaneously at other switching outputs. The sensor device 10 is configured to allow for multiple independent switching channels for the measured quantity and its derivative.

[0049] The time derivative of the measured quantity is determined by discrete differentiation, according to the formula: dMeasurement(t) / dt = (Measurement(t2)-Measurement(t1)) / (t2-t1).

[0050] The measured values ​​do not need to be taken at constant intervals. However, the respective measurement time must be known. Before dMeasurement(t) / dt is evaluated, this quantity can be low-pass filtered and / or averaged.

[0051] The decisive advantage over conventional, known methods is that by placing the filter F in front of the sensor, both a first target T1 and a further target Tn can be detected. This method makes it possible to monitor dynamic behavior such as velocity or acceleration. The sensor device 10 can distinguish between the approach and increase of the distance of target T1, Tn from the sensor element 20 by evaluating the sign of the derivative. The sensor device 10 determines the time derivative of the measured value P. The sensor device 10 can output this value as a measured value or, according to preset parameters, e.g., based on thresholds, logic, or delay times, determine a switching signal, output this signal, or control a switching output S1, Sn accordingly.The sensor device 10 according to the invention makes it possible to automatically detect several targets T1, Tn with only a single sensor element 20, thus simplifying such applications.

[0052] Fig. Figure 4 schematically shows a method for anomaly detection integrated into a sensor device 10. The sensor device 10 comprises a sensor element 20 for detecting objects T1, Tn, or at least one sensor element 20 for detecting a medium. The sensor device 10 according to the invention can detect either position or distance values ​​of an object or process values ​​of a medium. The method is based on the detection of a measured value, i.e., a measured quantity x(t), by a sensor element 20 integrated into the sensor device 10. This sensor element 20 is connected to an evaluation circuit AE, which processes the detected data. Parallel to and simultaneously with this detection of the primary measured value, the sensor device 10 performs additional measurements.These additional measurements include at least one of the following quantities: the first derivative of the measured value with respect to time, the second derivative of the measured value with respect to time, or the integral of the measured value over time. The parallel and simultaneous acquisition of these additional quantities enables a more comprehensive analysis of the measured value and its change over time. Within sensor unit 10, these acquired quantities are evaluated in parallel. This parallel evaluation ensures that all relevant data are processed simultaneously, thus increasing accuracy. A switching signal is generated for each of the evaluated quantities. This switching signal represents specific states or changes in the acquired data that may indicate potential anomalies. The evaluation circuit AE, which analyzes the data in real time, determines this switching signal.To detect an anomaly, the acquired switching information is combined using configurable operators. These operators are configured to logically link the various switching information, enabling a well-founded decision regarding the presence of an anomaly. The use of configurable operators allows the method to be flexibly adapted to different applications and specific requirements. The combination of parallel data acquisition, comprehensive evaluation, and the use of configurable operators ensures that anomalies can be detected reliably and efficiently. It is advantageous if the sensor device has multiple channels (channel 1, channel n) to simultaneously detect different conditions. This allows for more comprehensive monitoring and analysis of complex situations, as multiple parameters can be acquired and evaluated in parallel.This increases the accuracy and reliability of anomaly detection. Furthermore, the sensor device can be used more flexibly, as it is capable of monitoring different scenarios simultaneously. It is also advantageous if the means for logically linking the switching information include a parameterizable logic unit (PLU). Here, the parameters for the evaluations and the type of logical connections are determined through data analysis of the acquired measurements and derived values. This results in dynamic adaptability of the system, as it is able to automatically adjust to changing conditions. Moreover, the parameterizable logic unit (PLU) allows for individual configuration that can be tailored to specific requirements and applications.Additionally, integrating such a logic unit reduces the need for external, higher-level control units, which in turn lowers the complexity and cost of the overall system. A parameterizable logic unit (PLU) is a digital circuit that can be adapted via parameters. It allows the configuration of logical functions without physically modifying the hardware. Another advantage is that the parallel processing of data from multiple channels significantly reduces the system's response time. This is particularly important in time-critical applications, as potential anomalies can be detected more quickly and appropriate measures taken. Furthermore, using multiple channels enables redundancy in data acquisition, increasing the system's reliability. Should one channel fail or deliver faulty data, the other channels can continue to provide reliable information.Furthermore, using multiple channels allows for greater accuracy in data analysis. Different channels can be specialized for different types of sensor data, enabling more specific analysis. This leads to better differentiation between normal operating conditions and actual anomalies.

[0053] Fig. 5 shows the procedure according to Fig.4, with the difference that the evaluation in a sensor device 10 is fed to a neural network. Since the data analysis takes place directly in the sensor device 10, the raw data does not need to be transmitted to a higher-level control unit. This eliminates delays in data transmission and relieves the load on the higher-level control unit. The reduction in data transmissions also contributes to energy savings, which is particularly advantageous for battery-powered sensor devices. Neural networks are capable of recognizing complex patterns and relationships in the data, which further increases the accuracy and reliability of anomaly detection. Through continuous adaptation and improvement of the neural network, the system's performance can be further optimized over time. Reference symbol list 10 Sensor device 20 sensor elements AE Evaluation Unit F Filter P Measured value PLU parameterizable logic unit S1 first switching output Sn further switching output T1 first target Tn further target

Claims

[1] Sensor device (10) for anomaly detection comprising: - a sensor element (20) for detecting objects (T1, Tn) or for detecting a medium and an evaluation unit (AE) for generating corresponding measured values ​​of a measured quantity, in particular position, distance, or process values, - Means of recording these measurements and at least one of the following quantities: i. the first derivative with respect to the time of the measured value ii. the second derivative with respect to the time of the measurement iii. the integral over the time of the measurement, - Means for parallel evaluation of the recorded quantities, - Means to determine switching information for each evaluated quantity, - Means for the logical linking of switching information for anomaly detection. [2] Sensor device (10) according to claim 1, wherein the sensor device (10) has multiple channels to detect different conditions simultaneously, wherein the means for logically linking the switching information comprises a parameterizable logic unit (PLU) in which the parameters for the evaluations and the type of logical links are determined by a data analysis of the acquired measured quantities and derived quantities. [3] Sensor device (10) according to claim 1 or 2, wherein the sensor device (10) comprises a neural network as an evaluation unit (AE) to which the detected and derived quantities are fed in order to recognize complex situations. [4] Method for detecting anomalies with a sensor device (10) according to claims 1 to 3, comprising the steps: - Acquisition of a measured value by at least one sensor element (20) integrated into the sensor device (10) with an evaluation circuit (AE) - parallel acquisition by the sensor device of at least one of the following quantities (10): i. the first derivative with respect to the time of the measured value, ii. the second derivative with respect to the time of the measurement, iii. the integral over the time of the measurement, - parallel evaluation of the recorded quantities within the sensor device (10), - Determination of switching information for each evaluated quantity, - Linking the switching information using parameterizable operators to determine the presence of an anomaly. [5] Method according to claim 1, wherein the linking of the switching information is carried out using Boolean operators or in binary form. [6] Method according to one of the preceding claims, wherein the sensor device (10) outputs the measured values ​​and the derived quantities in addition to the switching information. [7] Method according to one of the preceding claims, wherein the parameters for the evaluations and the type of logical connections are determined by a data analysis of the recorded measured quantities, the derived quantities or the integral over the time of the measured value. [8] Method according to any of the preceding claims, wherein the sensor device (10) has multiple channels to simultaneously detect and output different conditions. [9] Method according to one of the preceding claims, wherein the derived quantities and the integral are additionally fed to a neural network implemented in the sensor device (10) in the evaluation unit (AE), wherein the data analysis takes place in the sensor device (10) so that the raw data does not have to be transferred to a higher-level control unit.

Images