Patent Application
20240353251
The invention relates to radar-based detection of a boundary layer on the basis of machine learning algorithms.
In automation technology, in particular for process automation, field devices by means of which various process variables can be detected are often used. The process variable to be specified can be, for example, a fill level, a flow, a pressure, the temperature, the pH value, the redox potential, a conductivity or the dielectric value of a medium in a process plant. In order to capture the corresponding measured values, the field devices each comprise suitable sensors or are based upon suitable measuring principles. A variety of various types of field devices is manufactured and marketed by the Endress+Hauser group of companies.
For measuring the fill level of products in containers, radar-based measuring methods have become established, since they are robust and low-maintenance. A key advantage of radar-based measuring methods is their ability to measure the fill level more or less continuously and without contact. In the context of this patent application, the terms “radar” and “high frequency” refer to electromagnetic waves with frequencies between 0.03 GHz and 300 GHz. Typical frequency bands at which fill-level measurement is performed are at 2 GHZ, 26 GHZ, 79 GHZ, or 120 GHz. The two common measurement principles here are the pulse transit time principle (also known by the term, “pulse radar”) and the FMCW (“frequency-modulated continuous wave”) principle. On the basis of the pulse transit time method and on the basis of the FMCW method, fill-level measuring devices can be implemented with comparatively low circuitry complexity and a high fill-level resolution in the sub-millimeter range. Radar-based measurement principles are described in greater detail in, for example, “Radar Level Detection, Peter Devine, 2000.”
In addition to freely-radiating radar measurement, in which the high-frequency signals are emitted or received via an antenna, the variant of guided radar also exists. In this case, instead of the antenna, an electrically conductive measuring sensor (for example, a coaxial waveguide or a metal rod) is used, which is lowered into the container in order to guide the high-frequency signals. Similarly to freely-radiating radar, the high-frequency signal in the measuring sensor is reflected at the fill level of the product surface and guided back along the measuring sensor towards the fill-level measuring device. This variant of radar-based, fill-level measurement is also known by the term “TDR” (time-domain reflectometry). This variant is advantageous in that, due to the guided signal radiation, less power is required for operating the fill-level measuring device. A TDR-based fill-level measuring device is described, for example, in US patent specification 10,07,743 B2.
Independently of the measuring principle (pulse transit time, FMCW, TDR), the fill-level value is specified by determining from the reflected received signal the signal maximum resulting from the product surface and its corresponding distance value. In general, the various radar-based measurement principles are described, for example, in “Radar Level Detection, Peter Devine, 2000.”
Depending on the type of product whose level is to be specified, the product can be pervaded with impurities, additives or foreign substances, which can form an additive product layer above or below the product, such as a foam, a sediment or an oily or aqueous phase. In this case, a more or less pronounced boundary layer, often also understood as a “separating layer,” is also formed between the actual product and the additive product layer. Accordingly, the fill-level measurement can be disturbed or falsified by possible additional product layers, since the corresponding signal maximum of the product surface cannot be identified with certainty. For this reason, it is of interest to be able to reliably detect not only the fill level of the product, but also the presence or even the thickness of an additive product layer. A TDR-based fill-level measuring device that determines a possible additive product layer by comparing the amplitude with a theoretical target value is described in European patent application 2 722 655 A1. However, even in this case, detection by means of identification of the corresponding signal maximum is at least uncertain, since the additive product layer or its surface may reflect the corresponding signal only very weakly.
For this reason, the invention is based on the object of being able to reliably determine a possible boundary layer.
The invention achieves this object by means of a corresponding measuring system for detecting a boundary layer on a product contained in a container. For this purpose, the measuring system comprises the following components:
Within the scope of the invention, the term “unit” in principle refers to any electronic circuit that is designed in a manner suitable for the intended purpose. Depending on the requirement, it can therefore be an analog circuit for generating or processing corresponding analog signals. However, it can also be a digital circuit such as a microcontroller or a storage medium operating in conjunction with a program. The program is designed to carry out the corresponding method steps or to apply the necessary computing operations for the respective unit. In this context, various electronic units of the measuring system in the sense of the invention can potentially also access a common physical memory or be operated physically by means of the same digital circuit.
The implementation of a machine learning algorithm according to the invention avoids the problem that the boundary layer or the additive product layer in the received signal often does not generate a clearly assignable signal maximum that traditional distance measuring methods according to the prior art, such as the TDR method, the pulse transit time method or the FMCW method, require for its detection. It is rather the case that received signals recorded under complex measurement conditions, such as multiple reflections, can be interpreted much better by means of machine learning algorithms.
In addition to the detection of the boundary layer according to the invention, it is also conceivable that the evaluation unit or the receiving unit of the measuring device is also designed to additionally determine the fill level of the product in the container with the aid of the received signal on the basis of traditional distance measuring methods.
In particular, within the scope of the invention, the evaluation unit can be designed to use the received signal to determine by means of the machine learning algorithm
In this case, the mass or volume fraction of the product in the boundary layer can be detected, for example, by the evaluation unit determining along the measuring sensor a distribution of the attenuation coefficient, of the conductivity and/or of the dielectric constant in the container by means of the machine learning algorithm.
Within the scope of the invention, a “machine learning algorithm” is defined in principle as any universal mathematical function approximator that maps at least one of its input variables to at least one of its output variables. Internal parameters of the corresponding mapping functions are adjusted during a teaching phase based on known training data. In this case, in particular, supervised learning is implemented as an algorithmic approach. In this connection, the specific form in which the machine learning algorithm is implemented is not firmly prescribed within the scope of the invention. For example, the machine learning algorithm can be implemented in the form of “decision trees,” a “support vector machine,” “naive Bayes classifiers,” or “k-nearest neighbor.” However, the boundary layer can be detected particularly effectively if the machine learning algorithm is designed on the basis of a non-symbolic approach, such as an artificial neural network, in particular in the form of a deep learning method. Machine learning algorithms are described in more detail, for example, in “Introduction to Artificial Intelligence” (Wolfgang Ertel, 2017).
Furthermore, how or where the evaluation unit is implemented is not essential within the scope of the invention. For example, the evaluation unit can be designed as an integral component of the measuring device, or as a component of a higher-level network, such as a cloud or server, or as a component of a plant-specific process control system.
It is advantageous if the measuring device is designed on the basis of the TDR method, such that the transmission unit is realized as a measuring sensor that extends into the container. Corresponding to this, the signal generation unit generates the high-frequency signal to be transmitted in this case according to the TDR method in a correspondingly pulse-wise manner. An advantage of the TDR method with respect to the idea according to the invention is that the received signal has an overall higher signal amplitude compared to freely-radiating radar, which in principle makes it easier for the machine learning algorithm to identify the boundary layer.
Corresponding to the measuring system according to the invention, the object underlying the invention is further achieved by a corresponding measuring method for detecting a boundary layer of a product in a container by means of the measuring system according to any one of the embodiments described above. In this case, the method comprises the following method steps:
The machine learning algorithm can be learned in the teaching phase, which is required before the actual measurement operation, for example by means of experimentally obtained and/or simulation-generated received signals, such as by means of “CST Microwave Studio.”
The invention will be explained in more detail with reference to the following figures. In the figures:
To understand the detection of a boundary layer according to the invention, a TDR-based measuring device 1 is shown in
According to the TDR method, the measuring device 1 comprises a measuring sensor 11 as the transmission unit 11 for high-frequency signals SHF, RHF in the direction of the product 2, 2′ or after reflection in the container 3. Contrary to the embodiment shown in
Via the measuring sensor 11, the high-frequency signal SHF to be emitted is conducted in pulsed form or with a frequency ramp in the direction of the product 2, 2′ according to the TDR method. Due to the jump in the dielectric value DK at the surface of the product 2, 2′, the emitted high-frequency signal SHF is then reflected at the level of the product surface in the measuring sensor 11 and received accordingly after a corresponding signal transit time t in the measuring device 1 as the received signal RHF. In this case, the signal transit time t of the signal SHF, RHF according to
depends on the distance d
from the container top to the product surface. In this case, c is the propagation speed of the high-frequency signal SHF, RHF along the measuring sensor 11, which lies within the range of the speed of light c.
To generate the high-frequency signal SHF, the fill-level measuring device 1 comprises a correspondingly designed signal generation unit—in the case of the TDR method, for example, this can be based on a capacitor that is correspondingly discharged to generate the pulse lasting approximately 100 ps to 1 ns. In the case of free-radiating radar according to the pulse transit time method or FMCW method, the signal generation unit can comprise, for example, a frequency-controlled high-frequency oscillating circuit or an oscillating crystal. In order for the signal generation unit to generate the high-frequency signal SHF according to the respective method at the required clock rate in pulse or ramp form, the capacitor or the crystal oscillator is driven in a correspondingly clocked or modulated manner.
Between the signal generation unit of the measuring device 1 and the measuring sensor 11, a transceiver switch is interposed in order to feed the received signal RHF after reflection in the container 3 to a receiving unit, in which the received signal RHF is digitized or recorded. In this case, the design of the transceiver switch is, in principle, not firmly specified. In the case of the TDR method, as is the case in the embodiment shown in
To determine the fill level L, the received signal RHF can be recorded in the receiving unit, for example, by undersampling the received signal RHF according to the pulse transit time principle such that the received signal RHF is stretched in time by a defined factor. In this case, the time prolongation factor depends upon the sampling rate. For this purpose, the corresponding sampling rate must be selected to achieve sufficient time prolongation, such that it differs from the clock rate of the emitted signal pulses SHF only in the per mille range. The time prolongation simplifies, from a circuitry perspective, the determination of the fill level L on the basis of the received signal RHF. In contrast to the embodiment shown in
The determination of the fill level L by means of the possibly time-prolonged received signal RHF is illustrated in more detail by the left-hand curve in
As shown in
The exemplary embodiment in
According to the invention, such parameters related to the boundary layer can be determined by applying a machine learning algorithm MLA to the possibly time-prolonged received signal RHF. In this case, which learning algorithm MLA is applied is not specified in principle. However, artificial neural networks, in particular “deep learning,” are proving to be particularly effective in this regard. This idea according to the invention avoids the problem that the boundary layer or the additive product layer often does not generate a clearly assignable signal maximum in the received signal, which would be required for its detection according to the prior art.
Since the application of the machine learning algorithm MLA may require a high computing power, it is advantageous if the machine learning algorithm MLA is not implemented in the measuring device 1 itself, but in the external evaluation unit 4, since the measuring device 1 may be subject under certain circumstances to a limited power supply in the process plant. In this case, the measuring device 1 or its receiving unit can transmit the recorded received signal RHF to the evaluation unit 4 via a suitable interface, such as “PROFIBUS,” “HART,” “Wireless HART,” “4-20 mA,” “Bluetooth” or “Ethernet.” The evaluation unit 4 together with the measuring device 1 thus forms a corresponding measuring system for detecting a boundary layer in the container 3 according to the invention. Provided that sufficient computing power is also available in the measuring device 1, the machine learning algorithm MLA for detecting a boundary layer can also be implemented in the measuring device 4 itself.
The application of machine learning algorithms MLA requires teaching the measuring system according to the invention under known conditions. For this reason, prior to regular measuring operation, the measuring device 1 is to be exposed to various teaching situations with known parameters, such as
The underlying received signals RHF can be obtained not only experimentally but also by means of simulation, such as by “CST Microwave Studio.” This avoids a time-consuming teaching phase in the process plant.
Instead of determining the respective fraction % indirectly via the dielectric value, it is also conceivable within the scope of the invention to determine the fraction % directly from the received signal RHF by means of the machine learning algorithm MLA, as is also illustrated in
With the exemplary embodiment shown in
| Number | Date | Country | Kind |
|---|---|---|---|
| 10 2021 115 871.0 | Jun 2021 | DE | national |
| Filing Document | Filing Date | Country | Kind |
|---|---|---|---|
| PCT/EP2022/063604 | 5/19/2022 | WO |
