Method for operating a sensor assembly and sensor assembly and device and device for processing data

Abstract

The invention relates to a method for operating a sensor assembly. The invention also relates to a device for processing data, which is suitable for implementing such a method. The invention further relates to a sensor assembly which is suitable for use in such a method and / or which interacts with such a device, and to a device having such a sensor assembly and / or such a device.

Description

Technical Field

[0001] The invention relates to a method for operating a sensor assembly. The invention also relates to a device for data processing, which is suitable for carrying out such a method. In addition, the invention relates to a sensor assembly, which is suitable for use in such a method and / or for cooperating with such a device, and to an apparatus having such a sensor assembly and / or such a device. Background Art

[0002] Processes, in particular processes in industrial environments, are increasingly dependent on the availability of sensor data. For example, corresponding sensor data can be recorded by means of sensors of a sensor assembly and used to control and / or regulate, for example, an apparatus such as a machine.

[0003] If invalid sensor data is processed due to damage to one or more sensors of such a sensor assembly, or if sensor data is not present at all, this can lead to severe restrictions on the process flow. In the worst case, this can lead to a shutdown of the entire plant or parts of the plant, with significant economic consequences.

[0004] Therefore, reliable operation of the sensor assembly is of vital importance. However, due to the numerous and diverse types of sensor assemblies and the increasing number of sensors per sensor assembly, ensuring reliable operation is becoming increasingly demanding. Summary of the invention

[0005] It is therefore an object of the present invention to overcome the disadvantages described in the prior art and in particular to specify means by which a sensor arrangement can be operated in a reliable and simple but cost-effective manner.

[0006] According to a first aspect, the invention achieves this object by a method for operating a sensor assembly having a plurality of sensors, wherein sensor data are received from each of the plurality of sensors as received sensor data of the respective sensor and are provided as source sensor data of the respective sensor for further data processing. If an error state is ascertained with respect to at least one particular sensor of the plurality of sensors, then substitute sensor data for the particular sensor is subsequently ascertained and provided as source sensor data of the particular sensor in place of the received sensor data of the particular sensor.

[0007] Alternative sensor data are ascertained with the aid of at least one trained first data model with machine learning support, wherein received sensor data and / or source sensor data of at least one sensor which is a non-specific sensor selected from a number of sensors as an auxiliary sensor and / or data based thereon are used as input data for the trained first data model for machine learning.

[0008] The present invention is therefore based on the surprising finding that a program that relies on sensor data of sensors of a sensor component can continue reliably even during a malfunction, a brief malfunction or a temporary disturbance of the sensors of the component and / or of the transmission channel for transmitting the sensor data, i.e. when the real sensor data of the sensors involved are at least temporarily replaced by synthetic sensor data.

[0009] Thus, the time until sensor data of a specific sensor are available again in a correct manner, ie, for example, the time interval until the sensor is repaired or replaced by a new sensor and / or until a transmission channel for the sensor data is functioning again in an error-free manner, can be reliably covered.

[0010] In this way, if the sensor data of the sensor in question are incorrect or even completely missing, the running program can continue without interruption or with only a very short interruption. The sensor assembly can thus be operated with increased reliability and the associated downtimes of programs that rely on the sensor data can thus be reliably avoided or at least reduced.

[0011] Surprisingly, it has been found that the synthesized data can be ascertained particularly reliably if the synthesized data are ascertained based on at least sensor data of at least some of the remaining sensors of the component in conjunction with a trained data model of machine learning.

[0012] In other words, the specific (defective) sensor is briefly replaced by a virtual sensor in the form of a trained data model for machine learning, and the synthetic data of this virtual sensor is provided for processing and used as a replacement for the real sensor data of the specific sensor.

[0013] To put it another way, in this regard, it is first recognized that the actual damaged sensor (specific sensor) is replaced by a virtual sensor, so to speak, so that it can continue to work in place of the actual sensor until the specific sensor is replaced or repaired. The virtual replacement sensor is then based on a machine learning data model ("machine learning model").

[0014] It has proven to be particularly advantageous that the proposed method can preferably be independent of the number of sensors of the assembly and the type of sensors of the assembly in principle. As a result, the proposed method can be used particularly flexibly for different sensor assemblies. In addition, the proposed method is also particularly easy to use in already existing sensor assemblies, because the physical sensor assembly does not need to be adapted for this purpose or only needs to be adapted within a controllable range. As a result, the method can also be used in an economically advantageous manner in combination with existing sensor assemblies. Thus, in this way, the advantages of the improved operation can be used in a variety of ways for existing and newly equipped sensor assemblies.

[0015] The proposed method can advantageously solve technical sensor faults of an apparatus, in particular a machine, because the program (or functionality or other broad entity) based on the sensor data can continue to run in the usual way. This can reliably avoid or at least reduce the risk of operating interruptions.

[0016] For example, in an apparatus with several sensors, for example four load cells, such as a platform scale, if one of the sensors (load cells) fails, the apparatus (platform scale) can continue to be used when operating according to the proposed method, because sensor data can still be provided due to the use of a virtual sensor as a substitute for the failed sensor. Thus, in the case of a platform scale, the total weight advantageously does not deviate from the true total weight by, for example, about 25% as in the conventional case.

[0017] It should be noted here that in the case of error-free operation of the sensor assembly, the received sensor data received by each sensor is preferably identical to the provided source sensor data of the corresponding sensor. However, if alternative sensor data for a particular sensor is ascertained due to an ascertained error state and provided as source sensor data for the particular sensor, the received sensor data received by the particular sensor (if any data is received) and the provided source sensor data of the particular sensor can and usually advantageously also be different.

[0018] Substitute sensor data may advantageously be provided as source sensor data for the specific sensor instead of the received sensor data of the specific sensor until the specific sensor again receives sensor data that complies with regulations, such as after replacement or repair of the specific sensor.

[0019] For example, invalid sensor data may be received from the sensor if the sensor is removed from the measuring location, for example if the sensor has been completely or partially removed and / or dropped.

[0020] It is particularly advantageous here to select a plurality or all of the sensors that are not specific sensors from the plurality of sensors as auxiliary sensors. This makes the database for determining the alternative sensor data more robust, since a plurality of sensors contribute to it.

[0021] Advantageously, the source sensor data of the auxiliary sensor, which are currently used as input data for the trained first data model for machine learning, have or represent the received sensor data currently received from the auxiliary sensor. This allows the current situation of the sensor assembly to be taken into account particularly reliably.

[0022] In one embodiment, a trained first data model of machine learning is provided after an error state is detected. Subsequently, it can be used to determine alternative sensor data.

[0023] In one embodiment, machine learning is continuously used to determine substitute sensor data, in particular substitute sensor data at least for a specific sensor, using a trained first data model. In this case, these substitute sensor data can then be provided as source sensor data instead of the sensor data of the specific sensor when a fault state is detected. In other words, the switch to substitute sensor data is only made when a fault state is detected. As a result, a series of source sensor data for a specific sensor can be provided without interruption or almost without interruption.

[0024] Preferably, error conditions associated with a particular sensor are ascertained at the first opportunity.

[0025] The proposed method is particularly advantageous for the operation of sensor assemblies of conveying devices, measuring devices, weighing devices, grinding devices, mixing devices, filtering devices, screening devices, drying devices and / or metering devices.

[0026] Preferably, the method is computer-implemented and / or implemented by means of a device for data processing that is configured to implement the method. The device for data processing will be further discussed below, wherein the description there will apply correspondingly here unless the context otherwise indicates.

[0027] In the sense of the present application, a sensor arrangement is preferably understood to be a set of at least two sensors, advantageously a set of sensors of the same type, wherein the sensor data are preferably evaluated and / or processed within the framework of a common data processing.

[0028] Advantageously, all sensors of the assembly are installed at a single device, for example in order to detect operating parameters and / or physical quantities of the device and / or its components.

[0029] In this application, unless the corresponding context indicates otherwise, the names "the xth data model trained by machine learning", "the xth trained data model of machine learning" and "the xth data model of the trained form of machine learning" (with "xth" used as the case may be, such as "first", "second" or "third") are used synonymously.

[0030] Alternatively or additionally, provision may also be made for a training of the first data model for machine learning to be carried out after the error state has been verified.

[0031] Using such newly trained data models, particularly convincing substitute sensor data can be generated, thereby enabling particularly reliable operation of the sensor assembly even in the event of a sensor failure or damage.

[0032] The training of the first data model for machine learning advantageously begins immediately or almost immediately after the error state is ascertained (i.e., approximately less than 1 minute, less than 30 seconds, less than 10 seconds, less than 5 seconds, or less than 1 second). As a result, the data model can be provided in a trained form particularly quickly and used to ascertain replacement sensor data. At the same time, sensor data that is as recent as possible can be used for training, which advantageously can contribute to a reliable data model.

[0033] However, it may also be preferred in embodiments to start the training after the error state has been verified with a delay and / or to add it to the task list of tasks to be completed. This makes it particularly advantageous to take into account the instantaneous load of the computer system responsible for the training in a variable manner. For example, the training can be started, in particular within a defined or definable time window, depending on the load of the system and / or the availability of resources.

[0034] In one embodiment, the training is performed with the aid of a device for data processing. A storage device may also be provided there, on which the trained data model is stored and / or from which the trained data model can be retrieved.

[0035] Alternatively or in addition, it can also be provided that the first data model for machine learning is trained or has been trained with the aid of at least historical sensor data of a specific sensor, in particular historical received sensor data, and / or historical sensor data of at least one sensor of at least one auxiliary sensor, preferably all sensors, in particular historical received sensor data.

[0036] It has been shown that replacement sensor data can be determined from the corresponding historical sensor data, in particular at least temporarily, reliably and in a particularly advantageous manner. This is particularly advantageous if the environmental conditions existing at the time of recording the historical sensor data have not changed in the meantime or have changed to a limited extent.

[0037] In order to train the model, in an advantageous embodiment, received sensor data of the component-specific sensor and the auxiliary sensor are used, for example, before the fault. It can thus be advantageous to a certain extent that the virtual sensor estimates its raw data from the raw data of the functioning sensor, in particular the selected auxiliary sensor (and thus estimates the raw data of the defective sensor).

[0038] The training of the first data model for machine learning can include, for example, using at least historical sensor data, in particular historical received sensor data, of at least one auxiliary sensor (or multiple auxiliary sensors) as input data for the data model and / or using historical sensor data, in particular historical received sensor data, of a specific sensor as real data. The input data and the real data are advantageously associated with each other.

[0039] Alternatively or additionally, provision may also be made for the training of the first data model for machine learning to show that a linear correlation exists between historical sensor data of the specific sensor on the one hand and historical sensor data of the auxiliary sensor on the other hand.

[0040] Linear association enables the use of particularly good and efficiently implementable data models.

[0041] For example, in a sensor arrangement with four sensors, at a certain point in time t, the linear correlation between the sensor data of the specific sensor S1 and the sensor data of the three auxiliary sensors S2 to S4 is advantageously described by the following relation:

[0042] S1(t)=a·S2(t)+b·S3(t)+c·S4(t)+d.

[0043] In the framework of training a machine-learned data model that implements this relationship, the coefficients a, b, c and d can then be determined in the present example, and a trained data model for machine learning can be provided after the training is complete. Thus, the currently received sensor data (S2 to S4) of the three auxiliary sensors can then be used in equations with the coefficients determined during the training with the aid of the trained data model, and the current replacement sensor data (S1) can be determined as a result.

[0044] In one embodiment, the data model is of the linear regression type, in particular of the stepwise linear regression type. Advantageously or alternatively, the most significant sensor of the several sensors (and its non-specific sensor) is selected as the auxiliary sensor.

[0045] Alternatively or additionally, provision may also be made for the historical sensor data of the specific sensor and the historical sensor data of the auxiliary sensor to be acquired within the same time window.

[0046] Preferably, the individual historical sensor data of the specific sensor and all auxiliary sensors are recorded at respectively the same time points, and preferably all time points are located within a time window. Thus, for example, each sensor records a measurement value per unit time (e.g. per second). Thus, there are measurement values ​​for each sensor per unit time. When the time window is 100 units of time long, the sensor data of each sensor also include 100 measurement values ​​in this case.

[0047] Alternatively or additionally, it can also be provided that the historical sensor data of a specific sensor and / or auxiliary sensor is received sensor data received by the corresponding sensor within a defined or definable time period before an error state is verified or data provided as source sensor data, wherein preferably (i) neither the specific sensor nor any of the auxiliary sensors is verified as having an error state within the time period, and / or (ii) a sensor among several sensors that is verified as having an error state within the time period is not selected as an auxiliary sensor.

[0048] For example, the time period can end immediately after the verification or already earlier, in particular with a defined or definable time offset.

[0049] The longer the time period is selected, the more information about the individual sensors and their behavior and properties is available as a database on the basis of which replacement sensor data can be determined and / or on the basis of which a data model can be trained. In the case of shorter time periods, replacement sensor data can advantageously be determined based on more recent sensor behavior. Thus, changes in sensor properties that occur over the long term, such as changes due to changing ambient conditions or progressive aging of the sensor, remain unaccounted for.

[0050] Advantageous time periods are 30 days or less, preferably 14 days or less, preferably 7 days or less, preferably 3 days or less, preferably 1 day or less, preferably 12 hours or less, preferably 6 hours or less, preferably 3 hours or less, preferably 1 hour or less, and / or 1 minute or more, preferably 1 hour or more, preferably 3 hours or more, preferably 6 hours or more, preferably 9 hours or more, preferably 12 hours or more, preferably 1 day or more, preferably 3 days or more, preferably 7 days or more, preferably 14 days or more, preferably 30 days or more.

[0051] By selecting only such sensors, for which no error condition has yet been determined, as auxiliary sensors, a particularly solid database can be provided for determining replacement sensor data, in particular for training data models.

[0052] Alternatively or additionally, it can also be provided that the sensor data, in particular historical sensor data, of a specific sensor are at least slightly correlated with the sensor data, in particular historical sensor data, of each auxiliary sensor.

[0053] This makes it possible in a particularly advantageous manner to ascertain meaningful substitute sensor data based on the sensor data of the auxiliary sensor.

[0054] In the sense of the present application, the data of the sensors are preferably at least slightly correlated if there is at least slightly a correlation between the data. For example, an existing correlation can be determined by a correlation coefficient of the sensor data.

[0055] For example, the correlation coefficient may take values ​​from -1 to 1. Advantageously, in the case of -1 there is a completely negative correlation between the sensor data, in the case of 0 there is no correlation between the sensor data (at least no linear correlation and therefore preferably no correlation in the sense of the present definition), and in the case of 1 there is a completely positive correlation between the sensor data.

[0056] In one embodiment, there is at least a weak correlation between the sensor data when there is a correlation coefficient between the corresponding sensor data, especially in terms of quantity of at least 0.1, preferably at least 0.2, preferably at least 0.3, preferably at least 0.4, preferably at least 0.5, preferably at least 0.6, preferably at least 0.7, preferably at least 0.8, preferably at least 0.9, preferably at least 0.95.

[0057] Alternatively or additionally, it may also be provided that an error state in a specific sensor is detected if

[0058] (i) at least temporarily no longer receiving sensor data from a particular sensor,

[0059] (ii) received sensor data or a statistical value thereof received from a particular sensor is above or below a defined or definable threshold value,

[0060] (iii) the received sensor data received from a particular sensor does not meet a defined or definable quality criterion

[0061] (iv) the results of testing the resistance of a particular sensor, particularly in the form of a load cell, indicate that the sensor is damaged;

[0062] and / or

[0063] (v) a value, in particular a maximum value, of a correlation between received received sensor data of the specific sensor and received received sensor data of at least one other sensor of the several sensors, in particular an auxiliary sensor, is above or below a defined or definable threshold value.

[0064] For example, a disturbance or failure of a transmission channel (e.g., a cable or a radio channel) may prevent the reception of sensor data from a particular sensor. This may occur in particular in the case of sensors from which sensor data is received via a data cable or via an air interface by radio.

[0065] For example, the statistical value may be an average value of the sensor data, in particular an average value over a defined or definable period of time.

[0066] The corresponding quality criterion can be, for example, the noise behavior of the sensor and / or the sensor data recorded therewith. It can therefore be advantageously provided that the noise behavior of the sensor and / or the sensor data is checked and the result of the check is preferably used as a quality criterion.

[0067] The resistance test can advantageously be carried out, in particular continuously or temporarily, with the aid of a testing device which is in operative connection with the sensor or is brought into operative connection and / or is comprised by the sensor.

[0068] For example, a correlation can be established between received sensor data received by a specific sensor and received sensor data received by each of two or more, preferably all other sensors among several sensors, in particular auxiliary sensors, and an error state can be verified when a defined or definable number of correlations (correspondingly between the sensor data of the specific sensor and the sensor data of the other sensors) obtains a value above or below a defined or definable threshold, in particular a maximum value.

[0069] Alternatively or additionally, it can also be provided that after verifying an error state related to a specific sensor, elimination of the error state related to the specific sensor is verified, and then received sensor data of the specific sensor is received again and / or provided as source sensor data of the specific sensor, and in particular alternative sensor data is no longer provided as source sensor data of the specific sensor.

[0070] As a result, a particularly reliable and automatic restoration to a normal operating mode is possible when sensors of the sensor arrangement, in particular specific sensors, again provide sensor data in a manner consistent with the specification, for example after a specific sensor has been replaced or repaired.

[0071] Preferably, elimination of the error state is checked temporally after checking the error state, in particular at a second point in time which is temporally after the first point in time.

[0072] In one embodiment, the ascertained replacement sensor data are compared continuously and / or repeatedly at intervals with the received sensor data received from the specific sensor, and based on the structure of the comparison, the elimination of the error state associated with the specific sensor is ascertained. Thus, the replacement sensor data provided by the corresponding data model can also be used for other purposes, namely to identify the operation of the specific sensor in a manner that complies with the regulations again. Normal operation can then be restored again, and the received sensor data received (from the specific sensor) can be provided again as source sensor data (of the specific sensor).

[0073] Alternatively or additionally, it may also be provided that sensor data is received continuously from each of the plurality of sensors, sensor data is received in parallel from all of the plurality of sensors, and / or result data of a trained first data model of machine learning is used as substitute sensor data.

[0074] Advantageously, when the sensor data of the sensor is continuously received, the sensor data is continuously received from the sensor. For example, the sensor data may be digital values ​​received at a certain cycle frequency.

[0075] Advantageously, when the sensor data of a plurality of sensors are received in parallel, the sensor data are received in parallel from the plurality of sensors via one or more transmission channels.

[0076] Alternatively or in addition, it can also be provided that a second data model in a trained form of machine learning is prepared at least for the specific sensor, and the ascertainment of the alternative sensor data includes, in particular, at least temporarily ascertaining the alternative sensor data with the aid of the second data model trained by machine learning prepared for the specific sensor, preferably at least until the training of the first data model trained by machine learning is completed, wherein the second data model trained by machine learning is preferably identical to the first data model trained by machine learning.

[0077] By using the trained second data model, replacement sensor data can be determined with the aid of the trained second data model after the error state has been verified without a large time offset (mainly in a preferred embodiment, without having to wait until the training of the first data model is completed, if necessary).

[0078] For example, the trained second data model can be used until the training of the first data model is completed, wherein after the training of the first data model is completed, the replacement sensor data are then ascertained with the aid of the trained first data model (and in particular no longer with the machine-learned trained second data model). As an alternative to this, the machine-learned second data model can be identical to the machine-learned first data model. Thus, the (first / second) data model for ascertaining the replacement sensor data is advantageously already available in a trained form when ascertaining the error state. This data model is then used to ascertain the replacement sensor data. It is then no longer necessary to modify the trained data model in order to ascertain the replacement sensor data.

[0079] Preferably, source sensor data and / or received sensor data of at least one of the auxiliary sensors, preferably all the auxiliary sensors, and / or data based thereon are at least partially used as input data for a trained second data model for machine learning, and / or result data of the trained second data model for machine learning are used as replacement sensor data.

[0080] Preferably, the embodiments made with reference to the training of the first data model are correspondingly applied in the training of the second data model. Therefore, in the training of the second data model, corresponding features can also be provided individually and in any combination. However, it should be understood that the reference point of the historical data is no longer the verification of the error state in this case, but is preferably determined by the start of the training of the second data model. That is, the received sensor data received before the training of the second data model or provided as source sensor data are advantageously understood in this case as historical sensor data of the corresponding sensor.

[0081] The descriptions made with reference to the ascertainment of substitute sensor data with the first data model preferably apply correspondingly with respect to the ascertainment of substitute sensor data with the second data model. Thus, corresponding features can also be provided individually and in any combination when ascertaining substitute sensor data with the second data model.

[0082] When the error state is detected, the second data model is advantageously already available in a trained form of machine learning.

[0083] Preferably, such a second data model trained for machine learning is prepared accordingly for each sensor of the component.

[0084] Alternatively or in addition, it can also be provided that a third data model in a machine-learned trained form is prepared at least for the specific sensor, the machine-learned trained third data model is preferably identical to the machine-learned trained second data model, and the machine-learned trained third data model is used to at least temporarily, preferably continuously, determine test sensor data for the specific sensor and compare them with sensor data received from the specific sensor, and an error state in the specific sensor is determined based on the result of the comparison.

[0085] By using the trained third data model, the virtual sensor can, as it were, continuously provide comparative sensor data, which can test the sensor data of the specific sensor. As a result, error states can be detected particularly reliably.

[0086] For example, the machine-learned third data model may be the same as the machine-learned first data model and / or the second data model.The (first / second / third) data model used to provide test sensor data may then seamlessly provide replacement sensor data starting from verifying the error state.

[0087] Preferably, source sensor data and / or received sensor data of at least one of the auxiliary sensors, preferably all the auxiliary sensors, and / or data based thereon are at least partially used as input data for a trained third data model for machine learning, and / or result data of the trained third data model for machine learning are used as test sensor data and / or replacement sensor data.

[0088] Preferably, the implementation scheme made with reference to the training of the first data model is correspondingly applied in the training of the third data model. Therefore, in the training of the third data model, corresponding features can also be provided individually and in any combination. However, it should be understood that the reference point of the historical data in this case is no longer the verification of the error state, but is preferably determined by the start of the training of the third data model. That is, the received sensor data received before the training of the third data model or provided as source sensor data are advantageously understood in this case as historical sensor data of the corresponding sensor.

[0089] Preferably, the embodiments described with reference to the ascertainment of substitute sensor data using the first data model apply correspondingly to the ascertainment of test sensor data and / or substitute sensor data using the third data model. Therefore, corresponding features can also be provided individually and in any combination when ascertaining test sensor data and / or substitute sensor data using the third data model.

[0090] Preferably, such a third data model trained for machine learning is prepared accordingly for each sensor of the component.

[0091] Alternatively or additionally, provision may also be made for the received sensor data and / or the source sensor data to be respectively assigned or assignable to individual sensors and / or for the source sensor data to be provided to a program, a module, a device and / or in the form of a control signal.

[0092] For example, the received sensor data and / or the source sensor data may also be present in a manner of being classified according to sensors, and / or the source of the source sensor data for each sensor may be known in other manners.

[0093] Providing source sensor data may generally include providing data to an entity (which may preferably be implemented in software, hardware or a combination of both). Thus, other entities, such as software and / or hardware modules, programs, functionalities, software functions and / or devices, such as equipment and machines, may also access the sensor data and / or receive the sensor data. When the source sensor data is provided as a control signal, it is particularly flexible. For example, such a control signal may be digital and / or analog in nature. The control signal may optionally have a plurality of sub-control signals, in particular as many sub-control signals as there are sensors in the component. For example, each sub-control signal may then represent the source sensor data of a single sensor of the component.

[0094] Alternatively or additionally, it can also be provided that the sensor assembly comprises two or more, in particular three or more, in particular four or more, in particular five or more, in particular six or more six, in particular seven or more seven, in particular eight or more eight, in particular nine or more nine, in particular ten or more ten sensors, and / or wherein all sensors of the assembly are of the same type.

[0095] For example, the sensors of the component, in particular the specific sensors and the auxiliary sensors, are all or at least partially current measuring sensors, voltage measuring sensors, force sensors, weighing sensors, acceleration sensors, motion sensors, speed sensors, rotational speed sensors, temperature sensors, ultrasonic sensors and / or eddy current sensors.

[0096] According to a second aspect, the invention achieves this object by proposing a device for data processing, which in particular has one or more interfaces for receiving sensor data from several sensors, wherein the device is suitable for implementing the method according to the first aspect of the invention.

[0097] The device for data processing can be implemented, for example, in software, hardware, or a combination of both. The device for data processing can alternatively or supplementarily have a memory (especially for storing the first data model, the second data model, and / or the third data model for machine learning), a processor, a receiving device, a transmitting device (for example, for transmitting source sensor data, especially control signals to an internal entity or an external entity) or any combination thereof. The device for data processing preferably has one or more interfaces for receiving sensor data from several sensors (especially several sensors of the sensor assembly used in the method according to the first aspect of the present invention).

[0098] According to a third aspect, the invention achieves this object by providing a sensor assembly, which in particular takes the form of several sensors, which are suitable for application in the method according to the first aspect of the invention and / or for cooperating with the device for data processing according to the second aspect of the invention.

[0099] Unless otherwise indicated in the context, the embodiments made with reference to the first aspect of the present invention also apply here with reference to the third aspect of the present invention. In particular, all advantages and features explained with reference to the sensor assembly used in the method according to the first aspect of the present invention also apply here correspondingly and completely. Therefore, reference is made to the above embodiments in this regard.

[0100] Therefore, in preferred embodiments, all features explained with reference to the sensor assembly used in the method according to the first aspect of the invention can also be provided in the sensor assembly according to the third aspect of the invention, both individually and in any combination.

[0101] According to a fourth aspect, the invention achieves this object by proposing a device, in particular a machine, which has a sensor component according to the third aspect of the invention arranged therein and / or has such a sensor component and / or has a device for data processing according to the second aspect of the invention.

[0102] Unless otherwise indicated in the context, the embodiments made with reference to the first and second aspects of the present invention also apply here with reference to the fourth aspect of the present invention. In particular, all advantages and features explained with reference to the sensor assembly used in the method according to the first aspect of the present invention and with reference to the sensor assembly of the third aspect of the present invention and alternatively or supplementarily with reference to the device for data processing according to the second aspect of the present invention also apply here correspondingly and completely. Therefore, reference is made to the above embodiments in this regard.

[0103] Therefore, in a preferred embodiment, all the features explained with reference to the sensor component used in the method according to the first aspect of the invention, with reference to the sensor component according to the third aspect of the invention and / or with reference to the device for data processing according to the second aspect of the invention can also be provided in a sensor component of the apparatus according to the fourth aspect of the invention and / or in an apparatus for data processing, individually and in any combination.

[0104] Preferably, the device is or has a conveying device, a measuring device, a weighing device, a grinding device, a mixing device, a filtering device, a screening device, a drying device and / or a metering device.

[0105] Alternatively or additionally, it can also be provided that (i) the device is or has a platform scale, and / or the sensor of the sensor assembly is a weighing sensor, (ii) the device is or has a screen, and / or the sensor of the sensor assembly is a motion sensor and / or an acceleration sensor, (iii) the device is or has a metering and dispensing device, and / or the sensor of the sensor assembly is a rotational speed sensor and / or a weighing sensor, (iv) the device is or has a belt scale, and / or the sensor of the sensor assembly is a weighing sensor, and / or (v) the device is or has a crane scale, and / or the sensor of the sensor assembly is a force sensor, in particular a weighing sensor. BRIEF DESCRIPTION OF THE DRAWINGS

[0106] Further features and advantages of the invention emerge from the following description, in which preferred embodiments of the invention are explained with the aid of the schematic drawings.

[0107] Here it is shown:

[0108] Figure 1 A schematic view showing a sensor assembly according to a third aspect of the invention together with an apparatus for data processing according to the second aspect of the invention;

[0109] Figure 2 shows a schematic view of an apparatus according to a fourth aspect of the invention;

[0110] Figure 3 A flow chart showing a method according to the first aspect of the invention; and

[0111] Figure 4 The course of the actual sensor data in comparison with the calculated substitute sensor data is shown. DETAILED DESCRIPTION

[0112] Figure 1 A schematic diagram of a sensor assembly 1 according to a third aspect of the present invention is shown, which is operatively connected to a device 3 for data processing according to the second aspect of the present invention.

[0113] The sensor assembly 1 has four identical sensors 5a to 5d, which are in the form of weighing sensors. Each sensor 5a to 5d is connected to an interface 9a to 9d of the device 3 via a corresponding transmission channel 7a to 7d, which are in the form of cables. The device 3 is configured to implement the method according to the first aspect of the invention.

[0114] The sensor assembly 1 can advantageously be used in a platform scale, for example by Figure 2 The apparatus 11 according to the fourth aspect of the invention schematically shown in FIG. 1 and together with the device 3 for data processing are implemented in Figure 2 The platform scale has a support surface 13 on which the object to be weighed can be placed and which is supported on the sensors 5a to 5d of the sensor assembly 1 (which are located on the support surface). Figure 2 When an object is placed on the support surface 13, a force will act on the sensors 5a to 5d due to the influence of the weight of the object. Each sensor 5a to 5d generates sensor data in the form of digital measured values ​​corresponding to the effect of the corresponding force.

[0115] Figure 3 A flow chart 100 showing a method according to a first aspect of the invention is shown.

[0116] In 101, during operation of the sensor assembly 1, sensor data of the individual sensors 5a to 5d are received by the device 3 via the transmission channels 7a to 7d as received sensor data and provided as source sensor data (e.g. to a program in the software). In the case of weighing sensors, the received sensor data will also change over time in accordance with the forces acting on the individual sensors changing in a time-dependent manner. In this case, current sensor data are received continuously and in parallel from all sensors. At least during the error-free operation of the sensor assembly 1, the received sensor data received by each sensor are advantageously identical to the provided source sensor data of the corresponding sensor.

[0117] Due to a failure of one of the sensors, such as the sensor 5a (which will be referred to as a specific sensor for better reference), sensor data are no longer received from the specific sensor 5a by the device 3 from a certain point in time. The failure of the sensor may be caused, for example, by damage within the specific sensor 5a or by a (physical) interruption of the transmission channel 7a (e.g. due to a cable break).

[0118] Due to the situation where sensor data is no longer being received from the specific sensor 5 a , an error state associated with the specific sensor 5 a is ascertained at 103 .

[0119] Next, in 105 , a first data model for machine learning is trained.

[0120] For this purpose, the remaining three sensors 5b to 5d of the component 1 are selected as auxiliary sensors, and their received sensor data of the last two days before the error state is verified are used as input data for training. At the same time, the received sensor data of the last two days before the error state is verified for the specific sensor 5a is used as real data associated with the input data for training. Therefore, the received sensor data of the last two days is historical sensor data here. Here, there is a linear correlation between the historical sensor data of the specific sensor 5a on the one hand and the historical sensor data of the auxiliary sensors 5b to 5d on the other hand.

[0121] It should be noted here that in the time period from which the historical sensor data of the sensors 5a to 5d originate, i.e. the last two days, no error state was detected for any of the sensors 5a to 5d. If an error state was detected, for example, a shorter and / or longer time period can be selected in which no error state was detected and the historical sensor data of the sensors 5a to 5d from this time period can be used.

[0122] Once the training of the first data model for machine learning is completed, the received sensor data of the auxiliary sensors 5b to 5d are used as input data for the first data model in 107. Here, the result data obtained by calculating the trained data model as the substitute sensor data is provided as the source sensor data of the specific sensor 5a. That is, the provided source sensor data of the specific sensor 5a is the ascertained substitute sensor data, while the provided source sensor data of the auxiliary sensors 5b to 5d is still the received sensor data received from these sensors 5b to 5d.

[0123] After a certain time, for example because a defective sensor 5a has been repaired or replaced, the sensor 5a is functioning in a prescribed manner again, so that sensor data are once again received by the device 3 from the specific sensor 5a.

[0124] Thus, the elimination of the error state associated with the specific sensor 5a is verified in 109. Next, the received sensor data of the specific sensor 5a are received again and provided as source sensor data of the specific sensor 5a. In particular, the substitute sensor data are no longer used. Thus, for example, the use of the trained first data model of the machine learning can then be stopped. The received received sensor data of each sensor are then again identical to the provided source sensor data of the corresponding sensor.

[0125] Figure 4The diagram shows the course of sensor data received by the actual sensor of the sensor arrangement according to the invention over a certain period of time (curve A) and the course of substitute sensor data for the same actual sensor calculated for this period of time with the aid of a trained data model for machine learning (curve B). The sensor arrangement has four sensors in this case. The data model used for machine learning is based on the historical sensor data of the four sensors in a manner similar to that of the machine learning in the reference period. Figure 3 The method described in the flowchart is used in a similar manner to the first data model used in the training. Figure 4 During the time period (time axis T) shown in the diagram, the trained data model for machine learning is calculated using the current sensor data of the remaining three sensors to determine the shown replacement sensor data for the sensor (curve B).

[0126] The almost identical course of the two curves A and B confirms the particularly advantageous and reliable way of operating the method proposed according to the first aspect of the invention for the operation of the sensor.

[0127] The features disclosed in the above description, the drawings and the claims may be essential to the invention in its various embodiments either individually or in any combination.

[0128] Description of Reference Numerals

[0129] 1 Sensor components

[0130] 3 Equipment for data processing

[0131] 5a Specific sensors

[0132] 5b, 5c, 5d Auxiliary sensors

[0133] 7a, 7b, 7c, 7d transmission channels

[0134] 9a, 9b, 9c, 9d interface

[0135] 11. Installation

[0136] 13 Support surface

[0137] 100 Flowchart

[0138] 101 receives sensor data from multiple sensors and uses it as each sensor

[0139] The source sensor data provided by

[0140] 103 Verify error status associated with a specific sensor among several sensors

[0141] 105 The first

[0142] A data model is trained

[0143] 107 Using the trained first data model of machine learning to identify alternative transmission

[0144] sensor data and replaces the sensor data of a specific sensor and provides the substitute sensor data as

[0145] Source sensor data for a specific sensor

[0146] 109 Verify that the error condition associated with the specific sensor has been resolved and reconnect.

[0147] Receive and provide sensor data of a specific sensor as source sensor data of a specific sensor

[0148] A, B Sensor data trends

[0149] X, T chart axes

Claims

1. A method for operating a sensor assembly having a plurality of sensors, comprising: receiving sensor data from each of the plurality of sensors as received sensor data for the corresponding sensor and providing the sensor data as source sensor data for the corresponding sensor for further data processing, wherein an error state is ascertained with respect to at least one particular sensor of the plurality of sensors, and subsequently ascertaining substitute sensor data for the particular sensor and providing the substitute sensor data in place of the received sensor data for the particular sensor as the source sensor data for the particular sensor, The alternative sensor data are ascertained with the aid of at least one trained first data model for machine learning, and received sensor data and / or source sensor data and / or data based thereon of at least one sensor other than the specific sensor selected from the several sensors as an auxiliary sensor are at least partially used as input data for the first trained data model for machine learning.

2. The method according to claim 1, wherein the first data model for machine learning is trained after the error state is verified.

3. A method according to any of the preceding claims, wherein the first data model for machine learning is trained or has been trained with the aid of at least historical sensor data of the specific sensor, in particular historical received sensor data, and / or historical sensor data of at least one sensor of at least one of the auxiliary sensors, preferably all sensors, in particular historical received sensor data.

4. A method according to any one of the preceding claims, wherein training of the first data model for machine learning shows that there is a linear correlation between the historical sensor data of the specific sensor on the one hand and the historical sensor data of the auxiliary sensor on the other hand. 5 . The method according to claim 3 , wherein the historical sensor data of the specific sensor and the historical sensor data of the auxiliary sensor are detected within a same time window.

6. A method according to any one of claims 3 to 5, wherein the historical sensor data of the specific sensor and / or the auxiliary sensor is received sensor data received within a defined or definable time period before the error state is verified or data of the corresponding sensor provided as source sensor data, wherein preferably (i) neither the specific sensor nor any of the auxiliary sensors is verified to be in an error state within the time period, and / or (ii) a sensor among the several sensors that is verified to be in an error state within the time period is not selected as an auxiliary sensor. 7 . The method according to claim 1 , wherein the sensor data, in particular the historical sensor data, of the specific sensor are at least slightly correlated with the sensor data, in particular historical sensor data, of each auxiliary sensor.

8. A method according to any one of the preceding claims, wherein an error state in the specific sensor is detected when: (i) at least temporarily ceasing to receive sensor data from the particular sensor, (ii) the received sensor data or a statistical value thereof received from the specific sensor is above or below a defined or definable threshold value, (iii) said received sensor data received from said particular sensor does not satisfy a defined or definable quality criterion, (iv) the result of testing the resistance of the particular sensor, in particular in the form of a load cell, indicates that the sensor is damaged; and / or (v) a value, in particular a maximum value, of the correlation between the received sensor data received by the specific sensor and the received sensor data received by at least one other sensor of the plurality of sensors, in particular an auxiliary sensor, is above or below a defined or definable threshold value.

9. A method according to any one of the preceding claims, wherein after verifying the error state related to the specific sensor, the elimination of the error state related to the specific sensor is verified, and then the received sensor data of the specific sensor is received again and / or provided as the source sensor data of the specific sensor, and in particular the alternative sensor data is no longer provided as the source sensor data of the specific sensor.

10. A method according to any one of the preceding claims, wherein the received sensor data is received continuously from each of the plurality of sensors, wherein the received sensor data is received in parallel from all of the plurality of sensors, and / or wherein result data of the first data model trained by machine learning is used as substitute sensor data.

11. A method according to any of the preceding claims, wherein a second data model in a trained form of machine learning is prepared at least for the specific sensor, and wherein the ascertainment of the alternative sensor data includes, in particular at least temporarily ascertaining the alternative sensor data with the aid of the second data model trained by machine learning prepared for the specific sensor, preferably at least until the training of the first data model trained by machine learning is completed, wherein the second data model trained by machine learning is preferably the same as the first data model trained by machine learning.

12. A method according to any one of the preceding claims, wherein a third data model in a form of machine learning training is prepared for at least the specific sensor, the third data model trained by machine learning is preferably the same as the second data model trained by machine learning, and wherein the third data model trained by machine learning is used to at least temporarily, preferably continuously, determine test sensor data for the specific sensor and compare it with the sensor data received from the specific sensor, and wherein an error state in the specific sensor is determined based on the result of the comparison.

13. A method according to any of the preceding claims, wherein the received sensor data and / or the source sensor data are respectively assigned or can be assigned to individual sensors, and / or the source sensor data are provided to a program, a module, a device and / or are provided in the form of a control signal.

14. A method according to any of the preceding claims, wherein the sensor assembly comprises two or more, in particular three or more, in particular four or more, in particular five or more, in particular six or more, in particular seven or more, in particular eight or more, in particular nine or more nine, in particular ten or more ten sensors, and / or wherein all sensors of the assembly are of the same type.

15. A device for data processing, in particular having one or more interfaces for receiving sensor data from a plurality of sensors, wherein the device is suitable for carrying out the method according to any of the preceding claims 1 to 14.

16. A sensor assembly, in particular in the form of several sensors, suitable for use in a method according to any of the preceding claims 1 to 14 and / or for co-operation with a device for data processing according to claim 15. 17 . An apparatus, in particular a machine, having a sensor arrangement according to claim 16 arranged thereon and / or having such a sensor arrangement and / or having a device for data processing according to claim 15 .

18. The device according to claim 17, wherein (i) the device is or has a platform scale, and / or the sensor of the sensor assembly is a weighing sensor, (ii) the device is or has a screen, and / or the sensor of the sensor assembly is a motion sensor and / or an acceleration sensor, (iii) the device is or has a metering and dispensing device, and / or the sensor of the sensor assembly is a rotation speed sensor and / or a weighing sensor, (iv) the device is or has a belt scale, and / or the sensor of the sensor assembly is a weighing sensor, and / or (v) the device is or has a crane scale, and / or the sensor of the sensor assembly is a force sensor, in particular a weighing sensor.

Images