Non-contact position / distance sensor with an artificial neural network and method for its operation

Abstract

In a non-contact position and / or distance sensor for determining the distance, spatial orientation, material properties or the like of a target object, and in a method for its operation, it is particularly provided that at least two sensor elements forming a sensor module (100) are provided and signals (103) supplied by the at least two sensor elements are jointly evaluated by means of at least one artificial neural network (110).

Description

State of the art

[0001] The invention relates to a non-contact position and / or distance sensor and a method for its operation according to the preamble of the independent claims.

[0002] In the field of measurement technology, non-contact position and distance sensors are known. For example, US patent 5,898,304 A1 discloses a corresponding sensor arrangement with an artificial neural network (ANN), which includes an electrical induction coil and an evaluation unit for the acquisition, processing, and evaluation of measured induction signals.

[0003] The KNN described therein comprises an input layer, at least one (hidden) intermediate layer, an output layer, and weights assigned at the junctions between any two individual layers. Suitable values ​​for the respective weighting factors are determined in a learning or training phase, during which test measurements are performed on a number of different target objects of known material and at known distances from the sensor. The sensor arrangement should be suitable for determining both distances and thicknesses, regardless of the material of the respective target object.

[0004] In the aforementioned displacement measurement system, the inductance signals or data measured by the measuring coil are subjected to spectral analysis using an artificial neural network (ANN). This analysis is based on the dependence of the measured spectrum on the spatial distance to the target object. The spectral analysis involves numerical calculations on a measured, time-varying quantity, in this case, an electrical voltage and an electrical current. However, these numerical calculations require considerable computational effort, thus preventing a compact design and cost-effective implementation of the sensor.

[0005] In the publication "ANN-based reduction for experimentally modeled sensors" by Pasquale Arpaia et al., published in "IEEE Transactions on Instrumentation and Measurement", 2002, Vol. 51, No. 1, pp. 23-30, a method for correcting the effects of multiple error sources in differential transducers is proposed. The correction is achieved using a nonlinear, multidimensional inverse model of the transducer based on an artificial neural network. The model utilizes independent information derived from the actual characteristics of the sensor elements as well as from an easily controllable auxiliary variable (e.g., the supply voltage of the evaluation circuit). Experimental results for correcting an eddy current displacement sensor subjected to combined disturbances from structural and geometric parameters are described. These results underscore the practical effectiveness of the proposed method.

[0006] EP 1 462 661 A1 discloses a position sensor for a fluidic cylinder-piston assembly, comprising a plurality of essentially identical sensor elements that cover a large detection range. In particular, the signals from the most relevant sensor elements are selected for evaluation and further processing. The sensor assembly includes a sensor array formed by Hall sensors arranged in a cylinder wall, comprising at least two Hall sensors spaced apart in the direction of piston movement, and a coil capable of carrying an electric current. Specifically, the switching points of the Hall sensors are adjustable, depending on the coil current, by means of the magnetic field generated by the coil.This approach not only allows for the electronic selection of individual Hall sensors, but also for interpolation between the individual Hall sensors, which allows the number of Hall sensors to be kept low while maintaining high precision in position detection.

[0007] German patent DE 197 51 853 A1 discloses a scanning element for an inductive position sensor with a segmented detection area, which also features sensor elements with the same or a similar measuring principle. The detection area consists of alternating electrically conductive and non-conductive segments, wherein the scanning element comprises a carrier element on which several excitation elements for generating a homogeneous electromagnetic excitation field and one or more flat sensor windings are arranged. In particular, two adjacent scanning tracks with sensor windings of different periodicities are arranged on the carrier element, and excitation elements are located laterally adjacent to each scanning track, so that a homogeneous electromagnetic excitation field is formed in the area of ​​the scanning tracks.

[0008] Furthermore, WO 94 / 08258 A1 describes a method for classifying the movements of objects along a passageway, in which a plurality of sensors arranged along the direction of movement of the objects and within the passageway are used to detect the objects. The classification of a segmented representation of corresponding passageway events is carried out using an artificial neural network. Disclosure of the invention

[0009] The invention is based on the objective of providing a position and / or distance sensor of the aforementioned type, which further develops the prior art concerned here.

[0010] The invention is based on the idea of ​​equipping a position and / or distance sensor with at least two sensor elements and jointly evaluating the measurement signals supplied by the at least two sensor elements using an artificial neural network.

[0011] A corresponding measuring unit comprises at least two sensor elements that can be influenced by the measured quantities and the possible measurement and environmental conditions. These sensor elements are based on physically different operating principles or on physically equivalent or similar measurement or operating principles, but preferably with different characteristic curves. A corresponding evaluation unit, comprising an artificial neural network, is preferably trained by a calibration process or by a teaching / learning process.

[0012] In the position and / or distance sensor according to the invention, it can be provided that the at least two sensor elements are operated statically or dynamically, e.g. in pulse mode.

[0013] In the aforementioned calibration or training process, an artificial neural network is preferably trained for each sensor element on the respective measured quantity under different conditions, so that the measured quantity can be determined independently of these conditions. This allows any inaccuracies of the sensor elements and / or the measurement electronics to be taken into account when creating a training data set, so that, due to the increased robustness of the signal evaluation according to the invention using an artificial neural network, corresponding manufacturing tolerances have essentially no influence on the subsequent measurements.

[0014] In well-known manufacturing processes, it can be disadvantageous if training the artificial neural network for a temperature-independent evaluation function requires recording the direct measurement signals of each sensor across a wide temperature range, particularly due to the prohibitively long time required for temperature-dependent signal acquisition. As an alternative to temperature-dependent training data acquisition, data can be acquired at the manufacturing temperature and the training dataset supplemented with simulated measurement signals for other temperature values.

[0015] For the reasons mentioned above, cost-effective sensor elements, such as those with magnetic field sensors or printed circuit board coils, can also be used. Furthermore, optimized printed circuit board technology for inductive and capacitive position and distance sensors can be employed, whereby manufacturing costs, signal strength, and any customer-specific limitations (e.g., housing, mounting dimensions, etc.) are considered as primary optimization parameters, in contrast to the significantly simpler evaluation of the primary measurement signals using electronic signal processing or equation-based mathematical analysis.

[0016] Furthermore, position / distance sensors can be implemented with a significantly larger ratio of possible measuring length to required sensor length compared to the state of the art, making almost any design possible, e.g. customer-specific designs, with virtually unlimited measuring capability.

[0017] The use of an artificial neural network also enables extended evaluation of the measurement results, including fault and function diagnostics, as well as the provision of a user-friendly setting aid for the purposes of presetting or calibrating the position / distance sensor.

[0018] Performing such a calibration or teaching / learning process has one or more of the following additional advantages, depending on the given measurement situation: - The aforementioned differences between the characteristic curves or (cross-)sensitivities of the individual sensor elements are factored out during the evaluation of the at least two sensor signals or are not included in the evaluation at all. - The material properties of the target object being measured also have no influence on the evaluation. - Possible irregularities in the arrangement of the sensor elements, both with regard to their possibly equidistant spacing in the longitudinal direction and deviations in the transverse direction, have no influence on the evaluation. The evaluation does not require an analytical or law-based relationship or evaluation algorithm between the actual measured quantity and the measured signals, nor between other quantities and the measured signals. This is because an artificial neural network can be trained using suitable training data, enabling the simulation or imitation of a large range of complex non-linear systems. The evaluation does not require a regular characteristic curve, i.e., a linear or non-linear relationship (1 / x, 1 / x). 2 , or sin / cos, etc.). - The sensor elements do not need to be shielded from environmental influences (e.g., installation, radiation, external magnetic fields, etc.). This is because if the aforementioned influences on the measured quantities or signals cannot be analytically separated, but the measurement signals contain / receive unambiguous information about the measured quantities, this information is reconstructed by a suitably trained KNN (k-nearest neighbors) algorithm. - The evaluation does not require any subsequent processing of the output signal from a corresponding evaluation unit. - The evaluation does not require either the uniformity or the similarity of the physical measurement principles of the individual sensor elements. - For the reasons mentioned above, a sensor with multiple sensor elements (“multi-element sensor”) can also be implemented, whereby the evaluation of the supplied sensor signals does not depend on knowledge of the relationship between the primary measurement signals of the individual sensor elements as well as the measured quantity to be determined and the aforementioned influencing factors. - In the case of such a multi-element sensor, it can also be designed as a "universal sensor" with several integrated measuring principles, whereby not only metals but also other materials such as plastics can be detected. - By using a correspondingly standardized platform that can be used for a wide variety of products (targets), the costs for the individual components of the position / distance sensor can also be reduced.

[0019] In the method according to the invention, it can further be provided that the at least one artificial neural network provides output signals which indicate the signal quality of the signals supplied by the least two sensor elements and / or the evaluation quality of the evaluation by the at least one artificial neural network. This makes the results supplied by the position and / or distance sensor according to the invention even more meaningful.

[0020] Finally, in the method according to the invention, the signals supplied by the at least two sensor elements can represent static or dynamic waveforms and serve as input signals for the at least one artificial neural network. An evaluation of such waveforms provides even more reliable position and / or distance results in an evaluation system or a position and / or distance sensor according to the invention. Brief description of the characters Fig. Figure 1 shows the basic structure of a position / distance sensor according to the invention, comprising an artificial neural network (ANN). Fig. Figure 2 shows the basic structure of an artificial neural network (ANN) according to the state of the art. Fig. Figure 3 shows an exemplary embodiment of a two-stage evaluation device of a position / distance sensor concerned here. Fig. Figures 4-22 show different application examples or embodiments of a position or distance sensor according to the invention. Detailed description of the exemplary implementations

[0021] From WO 2014 / 146623 A1, an inductive position / distance sensor, as described here, is disclosed. This sensor operates in a pulsed excitation mode, digitizing a transient induced voltage generated by a target object in a sensor element. The resulting digital signal is then evaluated using an artificial neural network (ANN). The ANN is trained to detect the distance or position of an approaching target object, regardless of its metallic composition. Alternatively, the patent provides for the detection or determination of the target object's composition or quality, independent of the actual distance.

[0022] A position / distance sensor according to the invention comprises several, in particular at least two, sensor elements, the signals of which are evaluated by a k-nearest neighbors (KNN) algorithm. These sensor elements either detect or sensing a matching physical quantity or different physical quantities. Alternatively or additionally, it can be provided that the measurement circumstances and / or environmental conditions present during the measurement, e.g., a metallic enclosure or housing, a metallic shield of a measuring device, or an ambient temperature, are also detected and evaluated accordingly. Again, alternatively or additionally, it can be provided that the physical principles used by the individual sensor elements are classified for evaluation purposes, e.g., with regard to their geometric structure or...the spatial arrangement of the sensor elements, the uniformity of the signals supplied by the sensor elements, as well as the sensitivity, crosstalk behavior, measurement resolution and the detectable measurement range of one or more sensor elements.

[0023] The following exemplary embodiments describe position and distance measurements in one, two, or three directions or dimensions, including the detection of rotational movements of a target object. Furthermore, the position / distance sensor described herein can also detect the spatial movement of a target object or simultaneously measure various physical properties or material dimensions of the target object.

[0024] The sensors described below according to the invention can generally be divided into the following three sensor types or sensor arrangements: 1. A one-, two- or three-dimensional arrangement (array) of essentially identical sensor elements, wherein the individual sensor elements are arranged along a sensitive axis, a sensitive circle or arc, within a limited area, in a limited volume or along an irregular surface or trajectory. 2. An arrangement (array) or group (cluster) of sensor elements operating according to the same or a similar physical principle, wherein the sensor elements have different measurement properties such as different sensitivity or cross-sensitivity to the primary measured quantity and, if applicable, to the aforementioned different measurement conditions and / or environmental properties. 3. An arrangement (array) or group (cluster) of sensor elements, each operating according to a different physical principle, wherein the sensor elements differ, for example, in their sensitivity or cross-sensitivity to the primary measured quantity and, if applicable, to the aforementioned different measurement conditions and / or environmental properties.

[0025] In a measuring arrangement according to the invention, subordinate measuring principles are distinguished from main measuring principles, e.g. in inductive measurements, self-inductive or transformer-like sensor elements are treated as not similar, or in the case of an optical main measuring principle, energy-based, flight-time measuring and / or triangulation measuring sensor elements, i.e., those operating according to different physical principles, are also taken into consideration.

[0026] It should also be noted that further sensor types can be derived by combining the three types mentioned. For example, an arrangement of type 1 can be formed from essentially identical sensor units, each of which has both a capacitive and an inductive sensor element. Such a sensor unit then provides two primary measurement signals.

[0027] It should also be noted that one or more of the sensor elements may additionally have a temperature transmitter or temperature sensor in order to detect a specified corresponding ambient temperature (as specified ambient condition).

[0028] The in Fig. Figure 1 shows the basic structure of a position / distance sensor according to the invention for detecting properties or parameters of a target object, such as the distance, position, or material of the target object. This structure initially comprises a preprocessing module 105, which is connected to a sensor module 100 via signal and data transmission. The preprocessing module 105 preprocesses the at least two primary or raw sensor signals 103 supplied by the sensor module 100. The sensor module 100 comprises at least two sensor elements (not shown) which supply the aforementioned at least two primary sensor signals 103. This preprocessing module also performs the necessary signal adjustments to feed the preprocessed signals 107 to an artificial neural network (ANN) 110.The output signals of the KNN 110 are fed to a post-processing module 115, by means of which the output signals 113 supplied by the KNN 110 are post-processed for the respective display by means of an output module 120, e.g. a screen module 120, and fed to the output module 120 as post-processed signals 117. By means of the output module 120, the level of the electrical voltage or the electrical current of the respective signals is adjusted, if necessary depending on the respective sensor type.

[0029] In the present embodiment, the sensor module 100 comprises a specific, predetermined number of sensor elements (not shown in this illustration) based on the same or different physical measurement principles. These measurement principles can be inductive, capacitive, optical, magnetic, magnetostrictive, or other physical operating principles. The sensor elements are arranged in predetermined geometric positions within the sensor module 100. Sensor elements based on different measurement principles can cooperate or interact in a suitable manner, as described below, to improve the detection accuracy or sensitivity of the position / distance sensor.

[0030] By means of the aforementioned signal preprocessing by the preprocessing module 105, normalization, amplification, scaling, digitization, reduction of the sampling rate, filtering, pre-evaluation, truncation (i.e., separation of signal components or corresponding signal shortening), or similar processes can be performed on the primary sensor signals 103. These processes generate or optimize the input data 107 required for the KNN 110 with the necessary data quality or data format. A suitable mathematical method familiar to those skilled in the art in the field of artificial neural networks can be applied in this signal preprocessing. As a result of the signal preprocessing, the number or content of the preprocessed signals 107 can differ significantly from the primary sensor signals 103.

[0031] The KNN 110 has one in the Fig. 2 input layer shown, to which the pre-processed sensor signals are fed, at least one (also in the Fig. 2 shown) hidden intermediate layer as well as a certain number of output neurons, by means of which the output data of the KNN 110 are generated.

[0032] The neural network is trained to extract 107 specific parameters from the input measurement signals, which correspond to corresponding parameters of the target object. These parameters include, for example, the distance between the sensor module 100 and the target object, the position of the target object above the sensor elements, the material of the target object, its mechanical density and / or surface properties, or similar characteristics. The KNN 110 converts the preprocessed sensor signals 107 into output signals or data 113 that correspond to the aforementioned properties of the target object.

[0033] By means of the aforementioned post-processing module 115, the output data 113 of the KNN 110 can be additionally filtered, truncated, interpolated and / or adapted to the required input signal of a aforementioned output module 120. For this purpose, the output module 120 includes the necessary devices for data conversion, signal amplification and / or adaptation.

[0034] It should be noted that a position / distance sensor designed according to the invention may require a more powerful microcontroller or a similar data acquisition and / or data processing unit compared to the prior art, due to the data processing using a KNN 110. Furthermore, the subsequent training and calibration of the KNN necessitates more powerful calibration devices with considerable computing capacity compared to the prior art.

[0035] The KNN 110 can be implemented as both a well-known "feedforward" network and a well-known "recurrent (neural) network". Fig. Figure 2 shows the typical structure of a feedforward KNN included here. The connecting lines between neurons of an input layer 200 and an intermediate layer 210 are shown as dashed lines for illustrative purposes only.

[0036] The KNN comprises an input layer 200, at least one hidden layer 210, and an output layer 220. Each input node or input neuron 201-205 of the input layer 200 is connected to each hidden neuron 211-216 located in the hidden layer 210 via predefined weighting factors 207. Each hidden neuron 211-216 located in the hidden layer 210 is connected to each output neuron 221 located in the output layer 220 via predefined weighting factors 207.

[0037] If the KNN has more than one hidden intermediate layer 210, then all input neurons 201 - 205 are connected to each neuron arranged in the first intermediate layer 210 via predetermined weight factors, wherein each neuron of a previous hidden intermediate layer is connected to each neuron of the following hidden intermediate layer via predetermined weight factors, and wherein all neurons of the last hidden intermediate layer are connected to each output neuron of the output layer 220.

[0038] Each neuron performs a summation of the values ​​provided by the preceding layer, each weighted with predefined factors, in a manner known per se, and evaluates the resulting sum using a neural function. A threshold value ("bias") can also be added to the input of each neuron during this summation. The result of the neural function's evaluation represents the output value of the respective neuron. Suitable neural functions include well-known functions such as a partially linear function, a sigma function, a hyperbolic tangent, or a sign function.

[0039] The single output neuron 221, located in output layer 220, provides the output values ​​for the entire KNN 110. Input layer 200 and output layer 220 are connected to the environment of the KNN 110, specifically to the preprocessing unit 105 and the postprocessing unit 115, respectively, whereas the aforementioned hidden layers or intermediate layers 210 are not directly accessible from the outside.

[0040] Fig. Figure 3 shows an embodiment of an evaluation unit according to the invention with a two-stage evaluation scheme in which two artificial neural networks KNN1 400 and KNN2 405 are arranged. In the present example, the individual sensor elements operate on the basis of the same physical principle and form a one- or multi-dimensional arrangement (array). It should be noted that a matching physical measurement principle is not a prerequisite for such an application of a two- or multi-stage evaluation scheme.

[0041] The primary sensor signals 410, the raw signals supplied by the sensor elements, are first fed to the input or an input layer (not shown here) of KNN1 400. KNN1 400 is trained to determine one or more influencing variables or parameters, e.g., the distance z required for linear position determination of the target object between the target object and the aforementioned arrangement of sensor elements (array). Alternatively or additionally, KNN1 400 can provide the data required for normalizing the primary sensor signals. The corresponding output data 415 of KNN1 400 are first fed to a rule-based preprocessing module 420 in this embodiment. The primary sensor signals 425 are also fed to the preprocessing module 420. The data, preprocessed as described above, are then fed to KNN2 405 or an input layer 430 (not shown).

[0042] In contrast to KNN1 400, KNN2 405 is trained to calculate and deliver the principal measurement quantities or parameters 435, as well as the position of the target object in relation to the aforementioned arrangement (array) of sensor elements.

[0043] It should be noted that the exact number of output nodes of KNN1 400 and especially KNN2 405 depends on the specific application. For example, KNN2 405 can provide position data about the target object along the x, y, and z axes in a three-dimensional or spatial arrangement of sensor elements.

[0044] The following describes a number of further application and embodiment examples of a position and distance sensor according to the invention. Example 1.1:

[0045] The in Fig. The position sensor shown in Figure 4a, equipped with a KNN (Kinetic Neighborhood Sensor), has an array 500 of, by way of example, six inductive sensor elements (SE1, SE2, ..., SE6) 505-530. The principal measured variable is the position of a target object 535 along an x-axis 540. The z-distance 545 is treated as the complementary measured value. Each of the sensor elements 505-530 is sensitive to electrically conductive and / or ferromagnetic target object materials. The amplitude of the oscillator ( Fig. 4b) changes with the movement of the target object in the x-direction. If the target object moves along the sensor arrangement 500 at a fixed z-distance 545, the resulting oscillator amplitudes (S1, S2, ..., S6) of the individual sensor elements 505 - 530 depend on the x-position of the target object ( Fig. 4b). This amplitude dependence is typically peaked, with the midpoint of the essentially symmetrical peak corresponding to the location of the respective sensor element. The maximum change in amplitude depends on the respective z-distance. A possible offset and adjustment of the individual sensor signals can be achieved using a (not shown) preprocessing stage. Example 1.2:

[0046] The in Fig. The sensor module shown in Figure 5a comprises a number of sensor elements (SE1, ..., SE4) 600-615, in this example four, configured as electrical measuring coils. An electrical voltage supplied by the measuring coils is detected, phase-sensitively demodulated 620-635, and fed into the input layer 640 of a (not shown) KNN. From these primary sensor signals, the current position of a target object 645 can be determined using the KNN. In addition to the array of sensor elements 600-615 shown, an array of, in this example, five electrical excitation coils (EC1, EC2, ..., EC5) 650-670 is also arranged, which are supplied, for example, by a harmonic voltage or current signal 675, such that at a given time, adjacent excitation coils always have opposite magnetic polarities. For a current orEach sensor element has a predefined z-distance. Fig. 5b shows a specific wave characteristic (S1, S2, S3, S4), where the wave amplitude depends on the current z-distance. Example 1.3:

[0047] The in Fig. The position / distance sensor shown in Figure 6a comprises a light-emitting diode (LED) 700, e.g., an LED operating in the visible or infrared frequency range, and a photodiode array 705. The LED 700 emits a focused beam of light 702, which forms an illumination spot 715 on the surface of a target object 710. In the present embodiment, the light 720 scattered back from the target object 710 is directed onto the photodiode array 705 by means of a Fresnel lens 725. The resulting intensity distribution at the photodiodes ( Fig. 6b) is a measure of the distance of the target object 710, in this case the primary measurement quantity. The intensity of the reflected light 720 additionally depends on the (surface) color as well as on the morphology or topology of the surface of the target object 710, which are processed here as complementary measurement quantities. In Fig. Figure 6b shows intensity distributions on the photodiode array 705 for two different object distances X1, X2, and for two different light scattering coefficients R1, R2. The measurement signals (S1, S2, S3, etc.) supplied by the photodiodes are read out by means of a control unit 730 and fed to the input layer 735 (of a KNN, not shown).

[0048] The use of a k-nearest neighbors (KNN) algorithm is particularly advantageous in this embodiment because neither the LED, nor the lens, nor the imaging geometry needs to be ideal. Furthermore, the entire compact sensor shown can be adapted to a customer-specific mounting geometry, and in particular, manufacturing costs can be significantly reduced.

[0049] Furthermore, reducing the number of input data points to the KNN can be advantageous as a preprocessing step for the primary measurement signals. For example, if the photodiode array has 128 diode elements, the number of input data points to the KNN can be reduced to just 16 by summing the intensity data from every 8 adjacent diode elements. Example 1.4:

[0050] The in Fig. The embodiment shown in Figure 7a comprises an array 830 of, by way of example, two inductive excitation elements 845, 850, preferably each operated with an alternating current, and six inductive receiver elements 800-825, which are designed as electrical coils provided with ferrite cores. The receiver elements 800-825 are sensitive to metallic target objects, the primary sensor signals of the individual receiver elements being, as shown in Figure 7a, Fig. As shown in Figure 7b, the x and y positions of a target object 835 depend on or correlate with these position data. The measurement signals also contain information about the distance z 840 between the arrangement 830 of sensor elements and the target object 835, which are processed here as complementary measured variables. A KNN (not shown) evaluates the x and y positions of the target object 835 and can provide the aforementioned z-information 840 within a limited measurement range. It should be noted that instead of the inductive sensor elements shown, which are based on the transformer principle, self-inductive or capacitive sensor elements can also be used. It should also be mentioned that the arrangement 830 of sensor elements 800–825 shown is extendable or expandable in both the x and y directions. Example 1.5:

[0051] At the in Fig. In the embodiment shown in Figure 8, it is assumed that the target object itself is formed by a rotating permanent magnet 900 and that analog magnetic field sensors 905-925 (e.g., Hall effect, AMR, or GMR-based) are arranged around this target object. Since the rotation of the magnetic target object 900 causes a periodic change in the output signals of the sensor elements, and the instantaneous rotation angle of the target object can therefore be derived from the sensor signals supplied to the input layer 930 of a (not shown) KNN in a known manner, a rotary encoder can be implemented with the arrangement shown. The KNN provides normalized sine / cosine output data. An incremental output signal, e.g., a high-resolution digital quadrature signal, can also be generated by means of a suitable post-processing step.The advantage of the KNN in this embodiment is that the positions of the magnetic field sensors 905 - 925 and the exact appearance or shape of the magnetic target object 900 do not have to be regular or ideal, thereby minimizing possible design limitations or boundary conditions in the development of a sensor concerned here. Example 2.1:

[0052] In the Fig. In the embodiments shown in Figures 9a-9c, two or more sensor coils 1000, 1005 are arranged in or around a ferromagnetic core 1010, each coil 1000, 1005 serving as the inductive part of a corresponding, alternately operated LC resonant circuit 1015, 1020. The resulting electrical oscillator voltages 1025, 1030 of the respective resonant circuit 1015, 1020 are sensitive to different amplitudes due to their different positions within the sensor head with respect to the distance to the target object and the metallic environment (so-called ambient conditions) of a (not shown) target object and / or sensor coils 1000, 1005. The recorded oscillator voltages 1025, 1030 are each demodulated 1035, 1040 and fed to the input layer 1045 of a (not shown) KNN, whereby the KNN is trained to calculate the target object distance from it, in particular independently of the aforementioned environmental conditions.

[0053] The in the Fig. The three different implementations shown in Figures 9a-9c relate to the following details: In example (a), the two coils 1000 and 1005 are arranged in two different opening areas or recesses 1050 and 1055 of the ferrite core 1010. In example (b), the two coils 1000 and 1005 are arranged within the same recess 1060 of the ferrite core 1010. In example (c), however, one of the two coils 1065 is arranged completely outside the ferrite core 1010 and serves as an inductive sensor element.

[0054] In addition to the Fig. The measurement curves shown in Figure 9a depict the distance-dependent oscillator amplitudes, i.e., the characteristic curves of the two embedded sensor units S1, S2, for two possible installation situations E1, E2, namely in the case of a measurement situation not installed (E1) and a sensor unit installed in a steel structure (E2).

[0055] The one in Fig. The electrical coils shown in Figures 9a-9c can also be used in a mixed operation, whereby one of the two coils forms part of an LC resonant circuit and is driven into oscillation by a suitable amplifier, while the other coil serves as the secondary coil of a transformer arrangement and receives a signal transmitted from the first coil. Both the oscillation amplitude of the LC resonant circuit and the phase and amplitude of the voltage induced in the receiving coil are influenced by the distance of the target object and the measurement conditions, e.g., the target object material, the mounting material, the mounting geometry, or the like. Example 2.2:

[0056] At the in Fig. In the embodiment shown in Figure 10, an inductive sensor coil 1105 with a ferrite core 1100 is used as part of an LC resonant circuit 1110. Parallel-connected capacitors 1115–1125, each with different capacitance values, are alternately connected by means of a low-impedance analog multiplexer 1130 to detect the resonant circuit amplitude for different frequencies. The corresponding demodulated amplitude values ​​1135, digitized by means of a conventional A / D converter (not shown here), are fed to the input layer 1140 of a k-nearest neighbors (KNN) processor (not shown). Since the influence of the aforementioned environmental conditions and the material of the target object also differs at different oscillation frequencies, the KNN can be trained to determine the distance to the target object independently of these conditions. Example 2.3:

[0057] The Fig. Figure 11 shows a "Type 2" sensor mentioned above, which has a coreless excitation coil 1200 by means of which the measuring system shown is excited or supplied with corresponding energy by means of a high-frequency magnetic field 1202. Furthermore, the system includes measuring coils 1205-1220 arranged at various positions, by means of which position-dependent information about a target object 1225 and its material properties is acquired. The signals 1230-1245 acquired by each sensor element 1205-1220 are first phase-sensitively demodulated 1250-1265. The resulting demodulated voltage signals are fed to the input layer 1270 of a k-nearest neighbors (KNN) (not shown), which is trained to calculate the distance between the sensors 1205-1220 and the target object 1225 from the voltage signals, independent of the material properties of the target object 1225.The measuring coils 1205–1220 can be arranged geometrically differently, each exhibiting different measurement sensitivities with respect to the target object 1225 and / or its material. A particularly advantageous arrangement can provide that pairs of measuring coils are connected in series in opposite directions. Since the majority of the voltages induced in the individual coil elements originate from the direct crosstalk of the excitation coil, the magnetic excitation signal emitted by the excitation coil 1200 can be suppressed even with non-optimal numbers of turns in the individual coil elements, while still achieving higher signal gains and better signal-to-noise ratios.

[0058] The one in Fig. The coil arrangement shown in Figure 11 can also be used in pulsed operation, namely when the excitation coil is operated with a current pulse instead of high-frequency excitation and the induced receiver signals are digitized and stored at a high sampling rate instead of the demodulation described, and selected data elements of the time-dependent waveforms are fed to the receiver coils of the input layer of the KNN. Example 2.4:

[0059] At the in Fig. In the embodiment shown in Figure 12, capacitive measuring sensor elements 1300-1310 of different sizes and geometries are provided, wherein the measured capacitances are digitized by means of a respective capacitive-to-digital converter (CDC) 1315-1325 and the sensor signals thus digitized are fed to the input layer 1330 of a k-nearest neighbors (KNN) algorithm (not shown). The sensor elements 1300-1310 exhibit different sensitivities to a target object (not shown) and, in particular, to the metallic environment, wherein the KNN algorithm is trained so that it can determine the distance between the sensor and the target object, independent of the environment. Example 2.5:

[0060] In Fig. Figure 13a shows an embodiment of a position-sensitive, multilayer (LE, LS1, ..., LS4) and essentially planar transducer 1400. The multilayer coil board shown has an excitation coil 1402 and several inductive receiver coils 1405–1420. The geometries of the receiver coils 1405–1420 are as shown in Fig. The coils shown in Figure 13a are not necessarily regularly designed, but can be optimized with respect to signal strength, signal-to-noise ratio, and / or manufacturing costs. With a metallic target object 1425 positioned above the transducer 1400, each individual receiver coil 1405–1420 provides different information or signals S1, S2, S3, S4, depending on the position 1430 of the target object 1425. Fig. 13b). Due to the aforementioned irregularity, this information cannot be evaluated analytically, i.e., using formulas. The advantage of using a KNN (not shown here) is that it can be trained to nevertheless determine or estimate the distance and / or position of the target object from this information.

[0061] It should be noted that even planar transducers of the present type, known in the prior art and possessing sinusoidal or even linear measurement characteristics, usually also exhibit disturbance properties in a real measurement situation, which means that the evaluation results must be subjected to a subsequent linearization process. Example 2.6:

[0062] At the in Fig. In the embodiment shown in Figure 14, the x and y positions of an optically opaque target object 1505 arranged in a slit or space 1500 of a support element 1502 are determined. An array 1510 of sequentially energized 1512 light-emitting diodes (LEDs) illuminates an array 1515 of photodiodes such that the optical shadow of the target object 1505 can be detected. The light intensity matrix 1520 supplied by all LED-photodiode pairs is acquired and fed to the input layer (not shown here) of a k-nearest neighbors (KNN) algorithm. The KNN is trained to determine the x and y positions of the target object from this information. The size of the target object can also be determined as a complementary measurement using this information. Example 3.1:

[0063] The in Fig. The multisensor shown in Figure 15 comprises a number of sensor elements 2000-2010, housed in a casing or sensor head 2002, which operate in completely different physical ways. In the present embodiment, these elements employ capacitive 2000, inductive 2005, optical (not shown here), and / or Hall effect-based 2010 measurement principles. A knnearest neighbors (KNN) processor with an input layer 215 (not shown) evaluates the measurement signals supplied by these sensor elements 2000-2010, which may be post-processed 2012, 2013 as described above, in order to detect the presence of any target object 2020-2030, and in particular regardless of its material. The target object could be, for example, a metal 2020, a non-metal such as plastic, a liquid 2030, or even a biological object 2025. Example 3.2:

[0064] The in Fig. The embodiment shown in Figure 16 relates to a position sensor formed from identical sensor units 2100-2120 operating according to the same physical measuring principle, wherein each sensor unit 2100-2120 comprises a capacitive sensor element 2125 (shown here only for sensor unit 2120) and an inductive sensor element 2130. The primary sensor signals supplied by the sensor elements are fed to the input layer 2135 of a k-nearest neighbors (KNN) algorithm (not shown). Since the inductive sensor elements 2130 are sensitive to both conductive and ferromagnetic materials, and the capacitive sensor elements 2125 are additionally sensitive to dielectric materials, a position sensor capable of detecting both metallic and non-metallic target objects 2140 can be implemented using the KNN algorithm.

[0065] It should be noted that in addition to the primary sensor signals mentioned, temperature data 2145 supplied by the sensor elements can also be taken into account during the evaluation by the KNN, as complementary sensor signals, so to speak. Example 3.3:

[0066] At the in Fig. In the embodiment and application example shown in Figure 17, an optical distance sensor 2200 is arranged in a window 2205, which has a capacitive sensor 2210 formed from an optically transparent, conductive oxide layer. The measurement signals supplied by the capacitive 2210 and optical 2200 sensor elements are fed to the input layer of a k-nearest neighbors (KNN) processor (not shown here). This type of combination of the aforementioned different measurement signals enables the reliable detection of various target objects with very different mechanical and / or electrical properties, as well as the effective suppression of interference signals. Example 3.4:

[0067] At the in Fig. In the embodiment shown in Figure 18a, a conventional inductive sensor coil 2305, provided with a ferrite core 2300, is used as part of an LC resonant circuit 2310. Both the demodulated voltage 2315 (see also Fig. 18b) as well as the frequency 2320 (see also Fig. 18c) of the resonant circuit 2310 are simultaneously acquired, and these two pieces of information are fed to the input layer 2325 of a (not shown) KNN as primary sensor signals. Based on this, a material-independent distance sensor can be implemented, since the target objects, which consist of different materials, exhibit significantly different and typical amplitude and frequency characteristic pairs depending on the distance of the respective target object. Example 3.6:

[0068] At the in Fig. In the embodiment shown in Figure 19, a sensor system comprising several measuring coils is formed by different layers 2500–2510 of a multilayer printed circuit board (PCB) arranged in a housing or sensor head 2502. The system operates according to two different measuring principles: an inductive measuring principle 2515, in which one or more of the aforementioned measuring coils 2500–2510 are used as receiver coils and in which a target object 2520 influences the voltages induced in the receiver coils; and a capacitive measuring principle 2525, in which the capacitance of one or more of the measuring coils 2500–2510 is detected. A control unit or...Measuring unit 2530 alternately performs measurements according to these two measuring principles 2515, 2525 by means of a conventional switching module 2535, whereby one or more inductive 2515 and one or more capacitive 2525 measuring signals are used as primary measuring signals and fed to the input layer 2540 of a (not shown) KNN in order to determine the target object distance and optionally, complementary measured quantities such as the target object material. Example 3.7:

[0069] At the in Fig. In the embodiment shown in Figure 20, measurement signals supplied by a known ultrasound-based distance sensor or transducer 2600 and by a capacitive sensor element 2605 are evaluated using a (not shown) k-nearest neighbors (KNN) algorithm with an input layer 2610. Such a measurement system advantageously enables the detection of target objects 2615 that cannot reflect ultrasound and / or of target objects that are already relatively close to, or even too close for a specific measurement principle, the sensor housing 2620.

[0070] Further embodiments are described below, in which an evaluation of two sensor signals as described above is carried out using sensor elements that are based on the same or at least similar measuring principle, but on different characteristic curves.

[0071] The Fig. Figures 21a to 21c show data measured on a linear measuring system equipped with six coils. The measuring system operates in a time-multiplexed mode, meaning that only one of the six coils is energized at any given time, with each coil operating in a self-inductive manner. The amplitudes of the respective oscillations serve as the sensor signals.

[0072] The in the Fig. The measurement results shown in sections 21a to 21c correspond to a raw signal ( Fig. 21b), a pre-processed signal ( Fig. 21c) and an output signal of the KNN ( Fig. 21a). In this embodiment, the KNN comprised nineteen hidden neurons and one output neuron, which were trained to determine the position of the target object. The measurement accuracy was tested using signal patterns corresponding to target object positions that were not used for training the KNN.

[0073] As from Fig. As can be seen in Figure 21a, the output signal of the KNN ("Network Output") plotted on the ordinate corresponds, within a range of approximately ±20 mm, to the position of the target object plotted on the abscissa around a reference position with the value 0, essentially to the plotted ideal curve ("Ideal Characteristics"). From the Fig. 21b further shows that the raw measurement signals of the individual coils represent downward-sloping bell curves 2700, 2705, which originate in the signal direction (ordinate) from slightly varying horizontal curves 2710, 2715 that are essentially horizontal at the baseline. As can be seen from Fig. As can further be seen in 21c, by a aforementioned preprocessing of the raw data 2700 - 2715, the bell curves 2700, 2705 are standardized in amplitude, specifically normalized to the value 1 in the present embodiment, and mirrored across the abscissa 2720, 2725 and the in Fig. The fluctuation of the horizontal lines 2710, 2715 running along the baseline shown in 21b is eliminated in 2730. It should be noted that the Fig. The normalization of the curves shown in 21c is only valid for a fixed, predetermined distance of the target object and therefore must be performed again for different distances.

[0074] Fig. Figure 22a shows an embodiment in which a position-sensitive sensor is formed by a multi-layer printed circuit board 2800 (“multi-layer PCB”). This type of sensor represents a transformer with a single input and multiple outputs. The bottom layer 2825 corresponds to an excitation coil, with the subsequent layers 2805–2820 representing sensing receiver coils. The electromagnetic penetration or coupling of the electric field excited by the excitation coil 2825 in the individual receiver coils 2805–2820 depends on the respective coil geometry of the receiver coils 2805–2820 and, in particular, on the position or distance of a target object, as described below. Fig. 22b can be seen.

[0075] As from the Fig. As can be further seen in Figure 22a, the receiver coils 2805–2820 consist of serially connected segments 2830–2855, each with alternating polarity. The layer 2820 following the excitation coil 2825 corresponds to a relatively long-wavelength sine signal, the subsequent layer 2815 to a similarly long-wavelength cosine signal, the layer 2810 following that to a relatively short-wavelength sine signal, and the uppermost layer 2805 again to a relatively short-wavelength cosine signal. Thus, there are two sine / cosine pairs, one with a relatively long-wavelength output and the other with a relatively short-wavelength output.

[0076] The dependence on the position of the target object results from the fact that each of the receiver coils 2805 - 2820 has a different sensor characteristic or characteristic curve, but all coils are subject to the same physical measuring principle.

[0077] The in Fig. 22b Measurement curves show two signal groups for the four receiver coils 2805–2820 mentioned above: two relatively long-wavelength sine / cosine pairs 2860, 2865, which provide a rough estimate of the target object's position, and two relatively short-wavelength sine / cosine pairs 2870, 2875, which enable a relatively accurate position determination. The short-wavelength data provide no information about the current wave period, whereas this period information is provided by the long-wavelength coil pairs 2815, 2820, since these only cover one wave period within the range shown in the diagram. Fig. The measuring range shown in 22c covers approximately 100 mm.

[0078] It should be noted that in Fig. Figure 22b shows further measurement curves, a total of eight curves, the remaining curves corresponding, among other things, to a quadrature phase. It should also be noted that the signal waveforms deviate significantly from ideal sine or cosine curves, with the exact signal waveforms depending strongly on the distance between the sensor and the target object.

[0079] The Fig. 22c to 22e show in a Fig. The evaluation data typically resulting from the position-sensitive sensor shown in 22a is shown. Fig. 22c shows the position output data supplied by a KNN as a function of the actual x-position of a target object, for different target object distances. Furthermore, it shows Fig. 22d Distance data supplied by a KNN as a function of the actual x-position of a target object, again for different target object distances e.g. Finally, the Fig. 22e Distance data supplied by a KNN as a function of the actual z-distance of a target object, measured at three different x-positions, namely x = 50, 95 and 140 mm. It should be noted that in the Fig. Examples 22c to 22e shown serve as input data for the KNN, each in Fig. The signal groups shown in 22b served this purpose.

Claims

[1] Non-contact position and / or distance sensor for determining the distance or spatial orientation of a target object, wherein at least two sensor elements forming a sensor module (100) are provided, which detect the position and / or distance of the target object, wherein the at least two sensor elements can be influenced by respective measured variables and possible measurement and / or environmental conditions, and sensor signals (103) supplied by the at least two sensor elements are jointly evaluated by means of at least one artificial neural network (110), wherein the sensor module (100) is connected via signal and / or data transmission to a preprocessing module (105) for preprocessing the at least two sensor signals (103) supplied by the sensor module (100), wherein the preprocessed sensor signals (107) are fed to the artificial neural network (110). characterized by, that the at least one artificial neural network (110) is trained by a calibration or learning process with respect to the sensor signals (103) supplied by the at least two sensor elements, specifically with respect to the respective measured quantity under different measurement and / or environmental conditions detected by means of an environmental sensor, and that the at least two sensor elements are based on physically different operating principles or on physically equivalent or similar operating principles and each exhibit different characteristic curves. [2] Position and / or distance sensor according to any of the preceding claims, characterized by that at least two sensor elements are operated statically or dynamically. [3] Position and / or distance sensor according to claim 2, characterized by , that the aforementioned dynamic operation corresponds to a pulsed operation. [4] Position and / or distance sensor according to any of the preceding claims, characterized by , that the output signals of the artificial neural network (110) are fed to a post-processing module (115), by means of which the output signals (113) supplied by the artificial neural network (110) are post-processed for the respective representation by means of an output module (120). [5] Position and / or distance sensor according to any of the preceding claims, characterized by , that two artificial neural networks (400, 405) are provided, wherein sensor signals (410) supplied by the sensor elements are fed to the first of the two artificial neural networks (400), the output data (415) of the first artificial neural network (400) are fed to a preprocessing module (420) and the data thus preprocessed are fed to the second of the two artificial neural networks (405) (430). [6] Position and / or distance sensor according to claim 5, characterized by , that the sensor signals (410) supplied by the sensor elements are additionally supplied to the preprocessing module (420) (425). [7] Method for operating a non-contact position and / or distance sensor according to any of the preceding claims, characterized by , that at least one artificial neural network (110) is trained by a calibration or learning process with respect to the sensor signals (103) supplied by the at least two sensor elements. [8] Method according to claim 7, characterized by , that in the calibration or learning process at least one artificial neural network (110) is trained for each of the at least two sensor elements on the respective measured quantity under different conditions. [9] Method according to claim 7 or 8, characterized by, that at least one artificial neural network (110) is trained to extract specific parameters from the pre-processed sensor signals (107) which correspond to corresponding parameters of the target object to be detected. [10] Method according to claim 9, characterized by , that the parameters mentioned relate to the distance between the sensor module (100) and the target object, the position of the target object above the sensor elements, or the surface properties of the target object. [11] Method according to any one of claims 7 to 10, characterized by , that at least one artificial neural network (110) converts the pre-processed sensor signals (107) into output signals (113) which correspond to the aforementioned properties of the target object. [12] Method according to any one of claims 7 to 11, characterized by, that at least one artificial neural network (110) provides output signals which indicate the signal quality of the sensor signals (103) supplied by the at least two sensor elements. [13] Method according to any one of claims 7 to 12, characterized by , that the sensor signals (103) supplied by the at least two sensor elements represent static or dynamic waveforms and serve as input signals for the at least one artificial neural network.

Images