Method for determining operating information of a metering pump

Abstract

Disclosed herein are embodiments of a method for determining operating information of a metering pump, the metering pump comprising a dosing chamber, a displacement member, and a drive motor for driving the displacement member, the method comprising receiving a plurality of detected values of an indicator quantity indicative of the strength of the actuation of the displacement member at each position of the displacement member during operation of the metering pump, and calculating the operating information from a machine learning model trained to output the operating information in response to receiving a plurality of input values derived from the detected values of the indicator quantity.

Description

Technical Field

[0001] The present invention relates to a method for determining the operating information of a metering pump.

Background Art

[0002] Metering pumps or dosing pumps are used to supply and dose precise amounts of liquid. These metering pumps typically have a movable displacement member in the form of a diaphragm or piston that is driven by a drive motor via a drive system that converts the rotational motion of the drive motor, for example, into the linear motion of the displacement member.

[0003] In many applications, it is desirable to determine the operating information of a metering pump, for example, to detect the presence of a malfunction or to calculate the effective stroke length or other operating parameters of the pump.

[0004] EP3591226 discloses a metering pump including a control device designed to detect the current position of a displacement element, detect the torque of an electric drive motor at several positions of the displacement element, and monitor the torque with respect to the position of the displacement element. This prior art metering pump includes an analysis module that compares the torque or pressure curves detected over time or compares the detected pressure or torque curves with previously stored sample curves.

[0005] However, there is still a desire to provide a method for determining the operating information of a metering pump that is applicable to different types of metering pumps or that can be at least efficiently adapted for use with different types of metering pumps. It is also generally desirable to provide such a method that can be configured in a highly reliable, computationally efficient, and cost-effective manner.

Summary of the Invention

[0006] Accordingly, there is still a desire to provide a method for determining operating information of a metering pump that solves one or more of the above problems and / or has other advantages or at least provides an alternative to existing solutions.

[0007] According to one aspect, embodiments of a method for determining operating information of a metering pump are disclosed herein, the metering pump comprising a dosing chamber, a displacement member, and a drive motor for driving the displacement member. Various embodiments of the method a) receiving a plurality of detected values of an indicator quantity indicative of the strength of the actuation of the displacement member at respective positions of the displacement member during operation of the metering pump, in particular a sequence of a plurality of detected values; b) calculating the operating information from a machine learning model that is specifically trained to output the operating information in response to receiving a plurality of input values derived from the detected values of the indicator quantity, in particular a sequence of input values; including.

[0008] The inventors have understood that by using a trained machine learning model, the operating information of the metering pump can be reliably determined from the detected values of the indicator quantity indicative of the strength of the actuation of the displacement member at each position of the displacement member during operation of the metering pump. Various embodiments of the method can be efficiently adapted for use with different types of metering pumps.

[0009] During operation of various embodiments of the metering pump, the drive motor preferably reciprocates the displacement member such that the displacement member increases or decreases the volume of the dosing chamber by its movement. When the dosing chamber is filled with an incompressible liquid, the change in the volume of the dosing chamber determines the volume of liquid supplied. In the presence of air or cavitation, the medium in the dosing chamber is compressible. In that case, the change in volume is different from the volume of liquid supplied. Generally, the supply capacity is the change in volume multiplied by the effective stroke length.

[0010] During operation of the metering pump, the pressure in the dosing chamber or a related indicator quantity, such as a related force or torque, is detected and recorded and can be used to determine the operating conditions or other operating information of the metering pump. The pressure or related indicator quantity can be measured continuously or intermittently, in particular periodically, as a sequence of detected values, in particular as a time series of detected values. The drive motor may be an electric drive motor.

[0011] The related force or torque may in particular be a force acting on the displacement member by the drive motor. The force or torque acting on the displacement member is related to, and in particular substantially proportional to, the pressure in the dosing chamber. Thus, some embodiments use the detected pressure in the dosing chamber as the indicator quantity, while other embodiments use the detected force or torque as the indicator quantity.

[0012] For the purpose of detecting the pressure, the metering pump can comprise a pressure sensor for detecting the pressure in the dosing chamber. In an alternative solution, the pressure can be calculated based on the drive torque or force provided by the drive motor. The calculation can be based on knowledge of the mechanical connection between the drive motor and the displacement member. The drive torque or force may, for example, be measured respectively by respective torque or force sensors or may be derived from the electrical values of the drive motor. Generally, further examples of indicator quantities can be derived from or otherwise related to the motor current, the motor voltage, or another quantity related to the motor current and / or the motor voltage and / or the motor load.

[0013] The inventors have found that a properly trained machine learning model can determine various operating information of a metering pump from a plurality of detected values of an indicator quantity. Examples of such operating information include the identification of discrete operating states as well as the prediction of continuous-valued operating parameters. Various embodiments of the methods disclosed herein enable the determination of such operating information without the need for specialized knowledge to identify how various operating states or parameter values can be derived from the measured indicator quantity. Accordingly, various embodiments of the methods disclosed herein use different types of machine learning models.

[0014] In particular, the operating information can include the classification of the operating conditions of the metering pump, particularly the classification of error conditions or malfunctions of the metering pump. The operating conditions can be current conditions or predicted future conditions, for example, even the prediction of imminent error conditions that are likely to occur in the near future. For this purpose, the machine learning model may include a classification model trained to output an identifier of one of a plurality of discrete classes. Examples of classes can include predetermined error conditions such as "cavitation", "bubbles", "leakage conditions", etc. In some embodiments, the classification model may be trained to output an estimated likelihood that one or more error conditions and / or other classes of operating conditions are present or will occur in the future, particularly in the near future.

[0015] Alternatively or additionally, in some embodiments, the operating information includes the values of operating parameters. Accordingly, the machine learning model may include a regression model trained to output the values of continuous-valued operating parameters. Examples of operating parameters include discharge pressure, effective stroke length, and discharge flow rate. For example, the value obtained by subtracting the effective stroke length from 1 is an indicator of the amount of air in the dosing chamber.

[0016] The machine learning model receives a plurality of input values representing the detected values of the indicator quantity and / or input values derived from the detected values of the indicator quantity. For this purpose, the machine learning model may be configured to receive different types of representations of the plurality of input values and / or additional input values. In some embodiments, the machine learning model is configured to receive a plurality of, for example, columns or arrays of input values of the indicator quantity, and each of the plurality of input values is associated with a respective position of the displacement member, and the machine learning model is configured to output the operation information in response to receiving at least the plurality of input values, for example, in the form of pairs of input data as described below, or in other forms. The input values may be detected values, or values derived from the detected values, for example, by a noise reduction process, averaging over a plurality of detected values, etc.

[0017] In some embodiments, for example, when the input values always represent the values of the indicator quantity at the same respective predetermined positions of the displacement member or at a respective predetermined time during the periodic movement of the displacement member, a one-dimensional representation of the indicator quantity may be used as input data for the machine learning model, and thus the one-dimensional representation of the indicator quantity enables a memory-efficient representation, which may be particularly beneficial for embedded implementations. Thus, in some embodiments, each of the plurality of input values is associated with a respective predetermined position of the displacement member and / or a respective predetermined time during the periodic movement of the displacement member. Thus, the input values represent a numerical sequence, and the numerical sequence has a predetermined phase relationship with the periodic movement of the displacement member.

[0018] In other embodiments, the machine learning model is configured to receive a plurality of pairs of input data, each pair of input data includes the position of the displacement member and the corresponding value of the indicator quantity at the position, and the machine learning model is configured to output the operation information in response to receiving the plurality of pairs of input data. Thus, the detected values at various positions can be used.

[0019] Thus, in the above and other embodiments, the machine learning model can receive a representation of a so-called pressure-stroke curve that associates a pressure or similar indicator quantity with the current position of the displacement member. The pressure-stroke diagram can be represented as a closed curve in a pressure-position coordinate system. Alternatively, the pressure-stroke diagram may be represented as an open curve that represents the pressure (or other indicator quantity) as a function of time or phase along the periodic movement of the displacement member.

[0020] In some embodiments, the machine learning model can receive a two-dimensional array of input values, where the two-dimensional array represents an array of image pixels that represents a pressure-stroke diagram, e.g., an image of a pressure-stroke curve. For example, the input may be represented as a raster image of the representation of the pressure-stroke diagram. Each pixel is represented as a number between 0 and 1 indicating the black level of the pixel. Other embodiments may use different representations of the pressure-stroke diagram, e.g., a more compact representation.

[0021] For this purpose, for the purpose of detecting the position of the displacement member, the metering pump can be provided with a suitable mechanism for detecting the current position of the displacement member. For example, the metering pump may include a position sensor, or the drive motor may be a stepper motor such that the position can be determined by counting the rotation angle of the drive motor.

[0022] Thus, in some embodiments, the method includes · receiving position data indicating the monitored position of the displacement member during operation of the metering pump, or at least calculating position data from the detected values of the received indicator quantity, and · calculating a plurality of input values, e.g., a column of values of the indicator quantity, or a plurality of pairs of input data, from the detected values of the received indicator quantity and from the received or calculated position data. and includes.

[0023] Some metering pumps enable determination of the position of a displacement member during operation, while other types of metering pumps do not provide this information. Thus, it is desirable to provide a method applicable to a wider range of metering pumps. For this purpose, in some embodiments, a machine learning model is configured to receive a time series of detected values of an indicator quantity at respective points in time and output the operation information in response to receiving the time series of detection positions of the detected values of the indicator quantity. In particular, in some embodiments, the machine learning model is configured to output the operation information based only on the received time series of detection positions of the detected values of the indicator quantity. It will be appreciated that when the cycle time of the movement of the displacement member, i.e., the stroke cycle, is known and the position of the displacement member is known over a reference time, the representation of the input data as a time series conveys the same information as the representation of the indicator quantity as a function of position. However, as described above, this information may not be readily available for all types of pumps. Nevertheless, the inventors have recognized that a properly trained machine learning model can determine useful operation information from only the indicator quantity, i.e., without the need to measure the position of the displacement member.

[0024] In particular, in some embodiments, the machine learning model includes a first machine learning model and a second machine learning model. The first machine learning model is configured to calculate an input representation of a plurality of input values of the indicator quantity based on the received time series of the detected values of the indicator quantity at each point in time during the operation of the metering pump. Each input value indicates the value of the indicator quantity at each respective, particularly at each respective predetermined or other known position of the displacement member. Thus, the second machine learning model can be configured to output operation information in response to receiving the calculated input representation. In particular, the first machine learning model may be trained to determine the phase and / or period of the periodic motion of the displacement member from the time series of the detected values of the indicator function. For this purpose, the first machine learning model may receive pressure values or values of another indicator quantity obtained during a time window that may have an unknown starting point with respect to the pump's stroke cycle and / or an unknown length with respect to the pump's stroke cycle. The first machine learning model can output the pressure value (or, if the first machine learning model receives a value of another indicator quantity, the value of said another indicator quantity) for a time window corresponding to one stroke of the pump, and the time window starts from a predetermined point of the stroke cycle, for example, the bottom dead center. The second machine learning model may then receive, as its input, the pressure (or other indicator quantity) value corresponding to a single stroke. Alternatively, the first machine learning model may output pressure values (or, if the first machine learning model receives a value of another indicator quantity, the value of said another indicator quantity) during a time window corresponding to a predetermined number of strokes or during a time window of a predetermined duration with respect to the duration of the stroke cycle. Thus, the second machine learning model can receive the pressure value or the value of another indicator quantity corresponding to the predetermined duration.

[0025] By providing separate first and second machine learning models, a more compact and memory - efficient representation of the overall model can be achieved. In particular, when a machine learning model includes, for example, multiple models for detecting respective operating conditions or for estimating respective parameters, the machine learning model can include a single first machine learning model that performs time - series phase detection and a plurality of second machine learning models that each receive the output of the first machine learning model as an input. However, in other embodiments, the first and second machine learning models may be combined into a single machine learning model, and thus, the combined machine learning model may receive a time - series of values of an indicator quantity, where the time - series is understood to have an unknown phase shift and / or an unknown duration with respect to the periodic motion of a displacement member. The combined machine learning model can be trained to directly output operation information from the time - series of the unknown phase and / or duration. Accordingly, a training set for such a composite machine learning model can include input time - series of different relative phase shifts and / or durations with respect to the periodic motion of a displacement member.

[0026] The present disclosure relates to various aspects, corresponding devices, systems, methods, and / or products including the methods above and below, each of which provides one or more of the benefits and advantages described in relation to one or more of the other aspects, and each of which has one or more embodiments corresponding to the embodiments described in relation to one or more of the other aspects and / or disclosed in the appended claims.

[0027] In particular, according to one aspect, disclosed herein is an embodiment of a computer - implemented method for creating a trained machine learning model. The method comprises a) Obtaining a set of training data items, where each training data item includes a plurality of input values and corresponding target outputs. The plurality of input values indicate an indicator quantity that represents the intensity of the operation of the displacement member of the metering pump at each position of the displacement member during the operation of the metering pump, and the corresponding target output indicates the operation information observable during the operation of the metering pump; b) Training a machine learning model from the obtained set of training data to output operation information in response to receiving the plurality of input values; including.

[0028] Generally, for the purposes of the present disclosure, the term "supervised machine learning model" refers to a machine learning model having a set of parameters such as weights adapted based on a set of training data using a suitable training algorithm such as an unsupervised or supervised training algorithm. Similarly, the term "training a machine learning model" refers to the process of adapting a machine learning model based on training data. In particular, the training, in particular the adaptation of the model parameters of the machine learning model, can be performed using supervised learning based on a set of training data in which each training data item is labeled with a corresponding target output.

[0029] Various embodiments of the methods disclosed herein can be implemented on a computer. Accordingly, disclosed herein are embodiments of a data processing system configured to perform the steps of the methods described herein. In particular, the data processing system may store program code adapted to cause the data processing system to perform the steps of the methods described herein when executed by the data processing system. The data processing system may be embodied as a single computer or other data processing device, or as a distributed system including a plurality of computers and / or other data processing devices, such as a client-server system, a cloud-based system, and the like. The data processing system may include a data storage device for storing computer programs and detector data. The data processing system may include a communication interface for receiving detected values and / or other types of sensor data. In some embodiments, the data processing system may be partially or fully embodied as a properly programmed or otherwise configured processing unit, such as a control device for controlling the operation of a metering pump. Accordingly, part or all of the data processing system can be housed within the housing of the metering pump, for example as part of a control device for controlling the operation of the metering pump, or as a separate processing unit. Alternatively or additionally, the data processing system can include one or more data processing devices external to the metering pump. The data processing system may receive detected values of an indicator quantity from the metering pump or otherwise, for example, from a separate pressure sensor.

[0030] According to one aspect, disclosed herein is an embodiment of a metering pump. Various embodiments of the metering pump include a displacement member, a drive motor for driving the displacement member, and a data processing system as disclosed above and below. In particular, the metering pump can be integrated with or separate from a control device configured to control the operation of the metering pump, and can include a processing unit, which can be configured to execute the steps of the methods described herein. The processing unit of the pump can operate alone as a stand-alone device or in cooperation with an external data processing system, such as a portable data processing device and / or a remote host computer and / or a cloud-based architecture, as part of a distributed data processing system, to execute embodiments of the processes described herein. The processing unit may be separate from the control device for controlling the operation of the pump, or may be partially or fully integrated with the control device. The pump may further include an integrated sensor configured to measure an indicator quantity or a quantity from which an indicator quantity can be derived.

[0031] According to another aspect, an embodiment of a system is disclosed herein, the system comprising a metering pump and a data processing system as disclosed herein, the metering pump comprising a dosing chamber, a displacement member, and a drive motor for driving the displacement member. In some embodiments, the data processing system is separate from the metering pump and includes an interface for receiving a plurality of detected values of an indicator quantity indicative of the intensity of the operation of the displacement member at respective positions of the displacement member during the operation of the metering pump. Alternatively, the metering pump includes the data processing system.

[0032] Yet another aspect disclosed herein relates to an embodiment of a computer program configured to cause a data processing system to perform the operations of the methods described above and below. The computer program may comprise program code means adapted to cause the data processing system to perform the operations of the methods disclosed above and below when the program code means are executed on the data processing system. The computer program may be stored on a computer-readable storage medium, particularly a non-transitory storage medium, or embodied as a data signal. The non-transitory storage medium may comprise any suitable circuitry or device for storing data, such as RAM, ROM, EPROM, EEPROM, flash memory, CDROM, DVD, magnetic or optical storage devices such as hard disks, and / or the like.

[0033] Preferred embodiments will be described in more detail in connection with the accompanying drawings.

Brief Description of the Drawings

[0034]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

Figure 8

Figure 9

Figure 10

Figure 11

Best Mode for Carrying Out the Invention

[0035] Detailed Description As an example of a dosing or metering pump, FIG. 1 schematically shows a diaphragm pump. It should be understood that the present invention can be similarly implemented with other types of dosing pumps, for example, a metering or dosing pump that uses a piston as a displacement member instead of a diaphragm. The pump shown in FIG. 1 has a pump or dosing chamber 2, the side walls of which are formed by a diaphragm 4. This diaphragm 4 is a displacement member. By displacing the diaphragm 4, the volume inside the dosing chamber 2 can be increased to fill the dosing chamber 2 and decreased to discharge liquid from the dosing chamber 2. A suction valve 6 is arranged below the dosing chamber 2, and a pressure valve 8 is arranged on the opposite side. Both valves are designed as check valves. In this example, the ball-shaped valve element closes the valve by gravity. However, a biasing element such as a spring can be further provided. During operation, liquid is sucked into the dosing chamber 2 from the liquid container 3 through the suction line 5 and through the suction valve 6, and is discharged from the dosing chamber 2 through the pressure valve 8. From the pressure valve 8, the liquid is discharged, for example, into the facility pipe 11 via the pressure line 9 and the pressure loading valve 7. The pressure loading valve 7 in the pressure line 9 determines the pressure in the pressure line 9, that is, maintains the pressure on the outlet side of the pressure valve 8 at a predetermined pressure. This pressure is set by the pressure loading valve 7. A pulsation damper 13 for equalizing the pressure pulsations generated in the outlet or pressure line 9 is connected to the supply line 9.

[0036] The diaphragm 4 reciprocates via the connecting rod 10. To reciprocally drive the connecting rod 10, a drive motor, in particular an electric drive unit in the form of an electric drive motor 12, for example a stepper motor, is provided. The rotary drive motor 12 moves the connecting rod 10 via an eccentric drive unit 14, converting the rotary motion into a linear reciprocating motion. The eccentric drive unit 14 is connected to the electric drive motor 12 via a gear drive unit 16. The connecting rod 10 is connected to the eccentric drive unit 14 at a connection point 18 which is offset from the axis of rotation x of the eccentric drive unit 14 by an eccentricity e. This causes a linear motion of the connecting rod 10 in the direction S when the eccentric drive unit 14 is rotated in the rotational direction R. In this example, furthermore, a spring 20 is arranged within the drive device. The spring 20 is a compression spring which is connected to the connecting rod 10 such that the spring 20 is compressed when the connecting rod 10 moves rearward in the direction S1, moving the diaphragm 4 to the retracted position. The spring 20 can accumulate energy during the suction stroke. This energy is released during the pressure stroke when the connecting rod 10 moves forward together with the diaphragm 4, i.e., to the advanced position in the direction S2. Thereby, the spring 20 smoothes the torque applied by the electric drive motor 12 over the entire stroke. It should be understood that it is also possible to arrange a spring which is compressed during the pressure stroke and acts as a return spring. Furthermore, the invention may be implemented without the spring 20.

[0037] The administration pump has a control device 22 that controls an electric drive motor 12. The control device 22 comprises a monitoring module 24 for monitoring the operation of the administration pump. The control device 22 can in particular comprise normal electronic components such as a CPU or other processing unit, a memory device, and one or more software applications for controlling the administration pump. The software applications can be stored on the memory device and be for execution on the CPU. The monitoring module 24 may preferably be implemented as a software module. In this example, the monitoring module 24 is integrated into the control device 22. However, it is possible to transfer information to an external computing or monitoring device, in particular a cloud device that functions as the monitoring module 24. For this purpose, the control device 22 can comprise a communication interface 26 for wired and / or wireless communication.

[0038] The monitoring module 24 is configured to continuously or intermittently record the pressure P in the dosing chamber 2 and the position of the displacement member. The pressure in the dosing chamber 2 and the position of the displacement member, for example, the position of the membrane 4 of the membrane pump in FIG. 1, can be recorded as the representation of the pressure-stroke curve in the pressure-stroke diagram. In this example, an encoder 28 for detecting the angular position of the rotor of the drive motor 12 is used to detect the position of the membrane 4 along the direction S. Furthermore, it is possible to detect a specific position of the drive or displacement member by, for example, a single sensor and calculate further positions based on the known speed and past time of the displacement member. Also, a stepper motor may be used instead of a dedicated encoder. Based on the knowledge of the transmission ratio of the gear drive 16 and the geometric design of the eccentric drive 14 based on the angular position, the position in the direction S can be calculated. The pressure P in the dosing chamber 2 may be detected by a pressure sensor 30, or indirectly detected by detecting the torque of the drive motor 12 or the force acting on the drive device and calculating the pressure P based on the force F acting on the membrane 4, or detected by another method. In this example, the pressure sensor 30 is arranged in the dosing chamber 2 and connected to the control device 22. When a force or torque is detected as the indicator quantity instead of the pressure, since the pressure is related to the force, in particular proportional to the force, or related to the torque, in particular related to the torque multiplied by a term depending on the position of the eccentric drive, it is possible to continuously record this force or torque over the position of the displacement member instead of recording the pressure.

[0039] The control device 22 further comprises a processing module 50 configured to implement a trained machine learning model. The processing module 50 may be implemented, for example, as a software module executed by a CPU of the control device 22, or in other ways. The trained machine learning module may include an appropriate representation of the model structure and parameter values of the model parameters, for example, the parameter values of the weights of a neural network model. In this example, the processing module 50 including the trained machine learning module is incorporated into the control device 22. However, it is possible to implement the processing module 50 and / or the machine learning module on an external data processing system outside the control device of the pump. For this purpose, the control device 22 may exchange information with an external data processing system, such as a cloud computing architecture, that functions as a processing module and / or implements the machine learning module. During operation, the processing module 50 receives the recorded pressure values and the associated positions of the displacement member from the monitoring module 24. The processing module 50 optionally processes the received information and supplies it to the trained machine learning module, which then returns the corresponding operation information as described herein. The processing module may be configured to display information on a display of the control unit and / or issue an alarm in case of a detected fault condition and / or transfer information and / or any issued alarm to an external device or system via, for example, the communication interface 26. Examples of the processing executed by the processing module 50 and the machine learning module are described in more detail below. The trained machine learning module may be commissioned together with the control device 22 during installation, or may be loaded into the control device 22 at a later time, for example, via the communication interface 26.

[0040] It will be understood that embodiments of the methods disclosed herein may also be implemented to calculate operating information of other types of metering pumps. The machine learning model may be incorporated into such a pump or implemented by an external computing device, for example, by a separate control unit communicatively coupled to sensors configured to measure the pump and / or indicator quantities. Further, the machine learning model may be implemented by a remote data processing system, such as a cloud service, configured to receive one or more indicator quantities from the pump and / or separate sensors and calculate operating information as described herein.

[0041] As further discussed below, some embodiments of the metering pump may not be able to monitor the position of the displacement member. Such pumps may only be able to monitor the pressure within the dosing chamber or another relevant indicator quantity. Thus, in such embodiments, the machine learning model may receive the pressure or other indicator quantity as its sole input.

[0042] FIG. 2 schematically shows an example of a pressure-stroke diagram depicting a pressure-stroke curve that can be detected, generally, by the monitoring module 24 or another method. The horizontal axis represents the stroke length S, i.e., the linear movement of the diaphragm 4 between the position representing the minimum volume of the dosing chamber 2 and the position defining the maximum volume of the dosing chamber 2, expressed as a percentage. The vertical axis represents the pressure P detected by the pressure sensor 30. 0 percent of the stroke corresponds to the bottom dead center 32 and 100 percent of the stroke length corresponds to the top dead center 34. The curve shows four stages of diaphragm movement. The lower part of the curve represents the suction stage 36, the left part with rapidly increasing pressure represents the compression phase 38, the upper part represents the discharge phase 40, and the right part with rapidly decreasing pressure represents the expansion phase 42 in which the internal pump volume expands. The expansion phase 42 corresponds to the movement of the diaphragm 4 in the direction S1, together with the suction stage 36, while the compression phase 38 and the discharge phase 40 form the pressure stroke in the direction S2.

[0043] When the monitoring module 24 of the control device 22 records or monitors the pressure and related displacement values continuously or intermittently, the monitoring device can detect changes in the pressure-stroke curve over time or over several strokes. Different problems or malfunctions that can occur in the dosing pump affect the course of the curve in the pressure-stroke diagram differently.

[0044] Figure 3 shows examples of the pressure-stroke curves of a dosing pump in the presence of different malfunctions or other operating conditions such as cavitation, the presence of air bubbles. Other indicator quantities such as force or torque can be represented in a similar way depending on the stroke length and can thus be related to different forms of indicator-stroke diagrams that can be used to detect the operating conditions and estimate the operating parameters in a similar way.

[0045] The various embodiments of the method disclosed herein provide an efficient way to reliably detect such problems or malfunctions from the recorded pressure, or other indicator quantity, and displacement values.

[0046] Figure 4 schematically shows a flowchart of a training method for creating a trained machine learning model for subsequent use in a method for determining the operating information of a metering pump.

[0047] First, the process obtains a set of training data items. For this purpose, in step S1, the process obtains a set of input columns. Each input column represents a value of an indicator quantity indicating the intensity of the operation of the displacement member of the metering pump at each position of the displacement member during the operation of the metering pump. For the purposes of the following description, the embodiments of the method and apparatus disclosed herein are mainly described with reference to pressure as the indicator quantity. However, it is understood that various methods and apparatuses can use other indicator quantities instead of or in addition to pressure.

[0048] The input column can represent a time series of measured pressure values P(t0), P(t1),..., P(t n ), where the times t0,..., t n are each time during at least one stroke cycle of the metering pump. The number of recorded values n (n>1) can depend on the speed at which the metering pump records values. Alternatively, the input column can be a column of pairs of pressure and position data: (P1, S1), (P2, S2),..., (P n , S n ). Here, each pair (P i , S i ) represents the pressure and the corresponding position, in particular the position of the displacement member at which the pressure is measured. For example, it will be understood that the position S i need not be explicitly included in the input column if the measured pressure values always represent the pressure values at respective predetermined positions. Instead, the input column may be represented as a pressure vector p[i], i = 1,..., n, where the index i enumerates the predetermined positions S i . It will be further understood that other embodiments may use other representations of the input column, for example, using measured torque or force instead of pressure. In one embodiment, the input column can be represented as an array of value / data pairs. In other embodiments, the input column may be represented in different ways, for example, as an image of a pressure-stroke curve, where the pressure-stroke curve represents a column of pairs of pressure and associated position data.

[0049] The input column may be obtained by operating a plurality of metering pumps under various operating conditions and by recording pressure and / or position data from the corresponding sensors of the pumps. For example, the recorded pressure and / or position data can be obtained by a metering pump monitoring module as described above in connection with FIG. 1, or by other means. The recorded data may be received from the pump via a suitable communication interface of the pump, or they may be stored locally and later read from the pump's data storage device or a separate monitoring unit.

[0050] In step S2, each of the acquired input sequences is labeled with one or more target output values indicating the operating information of the pump operating when the input sequence is acquired.

[0051] The target output values can be obtained in various ways, for example, manually determining operating conditions of the pump, such as malfunctions, and assigning a class identifier representing the determined operating conditions to the target values. For example, some operating conditions such as cavitation can be induced by, for example, restricting the pump inlet. In other embodiments, the target values can be obtained by, for example, performing a reference measurement of the output pressure or effective stroke length of a metering pump. For example, the target value of the effective stroke length can be determined from the flow rate measurement as well as the pump speed and pump volume.

[0052] In any case, the process can store the acquired target output values in association with the corresponding input sequences to which they are related, thus creating a set of labeled training data items. Each training data item includes an input sequence and a corresponding target output. The input sequence indicates an indicator quantity indicating the intensity of the operation of the displacement member of the metering pump at each position of the displacement member during the operation of the metering pump, and the corresponding target output indicates the observable operating information during the operation of the metering pump. It will be understood that the training data can include multiple target values for each input sequence, for example, an identifier for classifying operating conditions and / or respective values of one or more operating parameters. Thus, the training data can be used to train a single machine learning model or different machine learning models to output respective types of operating information.

[0053] In step S3, the acquired training data is used to train a machine learning model to output operating information in response to receiving a plurality of input values, particularly in response to receiving an input sequence of input values. The machine learning model may be a feed-forward neural network such as a convolutional neural network.

[0054] For example, a convolutional neural network can receive an input that represents a pressure-stroke curve. The pressure-stroke curve may be represented as a sequence of 1D data points, as a sequence of 2D data points, or as a two-dimensional input representation of a pressure-stroke diagram depicting the pressure-stroke curve, or otherwise. In this case, the machine learning model may be a feedforward neural network such as a convolutional neural network.

[0055] Here, when the input values are represented in a time series, the machine learning model can receive a one-dimensional sequence of the input. In this case, the machine learning model can be a feedforward neural network, a recurrent neural network, or another suitable type of machine learning model.

[0056] Generally, when the input to the machine learning model is a sequence or a one-dimensional array, the network can be a dense fully-connected network, a one-dimensional convolutional network, and / or a recurring network such as a long short-term memory (LSTM) network, or a combination thereof. For example, the first layer of the network may be convolutional, and the following layers may be fully connected. In one example, the convolutional and LSTM layers can be combined such that, for example, the first layer is a 1D convolution, the subsequent layer is an LSTM, and further layers are dense.

[0057] The training of the machine learning model can be based on a suitable training algorithm known per se in the art, for example, an algorithm based on backpropagation.

[0058] The training process results in a trained machine learning model, which can then be used to predict operation information regarding a metering pump in response to receiving a plurality of input values, as described herein. For this purpose, the representation of the trained machine learning model can be stored within the processing or control device of the metering pump or on another data processing system external to the pump, for example as a computer program module. For example, the representation of the trained machine learning model can be loaded as a processing module 50 onto the metering pump as described in connection with FIG. 1 above or otherwise described, or implemented in another way.

[0059] FIG. 5 schematically shows a flowchart of a method for determining operation information of a metering pump.

[0060] In a first step S4, the process records pressure and / or position data from the corresponding sensors of the metering pump for which the operation information is to be determined. The process may record the data during normal operation of the pump and represent the recorded data as an input sequence of the recorded values or in another suitable form. In other embodiments, it will be understood that the process may obtain detected values of another type of indicator quantity indicative of the strength of actuation of the displacement member at each position of the displacement member during operation of the metering pump. Further, in some embodiments, the process may obtain only the values of the indicator function, and in other embodiments, the process may record the values of the indicator function as well as the corresponding position data. The process supplies the obtained data to a suitable processing unit that receives the indicator function values and optionally the corresponding position data as inputs.

[0061] In subsequent step S5, the process derives a plurality of input values, particularly a sequence of input values, from the detected value of the indicator amount, and supplies the input values to a trained machine learning model trained according to the process described in relation to FIG. 4, for example. For example, deriving the input values may involve directly using the detected value, or may involve other preprocessing steps such as suitable scaling and / or suitable filtering of the input values. The particular preprocessing steps may depend on whether the machine learning model is configured to receive a representation of the input values, for example, a 2D representation of a pressure-stroke curve, a time series of optionally normalized pressure values, etc., trained machine learning model.

[0062] In response to receiving the sequence of input values, the trained machine learning model outputs one or more corresponding output values. Depending on the type of operation information represented by the output of the trained machine learning model, the output values may have different forms. For example, if the operation information represents a type of operation condition, such as a type of failure, the machine learning model may be a classification model, and the output value may represent a class identifier representing the operation condition of the metering pump. If the operation information represents a continuous quantity, the trained machine learning model may be a regression model, and the output value may be a continuous value representing the continuous quantity. In some embodiments, the machine learning model may output a plurality of output values. For this purpose, the machine learning model may include a plurality of submodels each trained to output a particular type of operation information. Different submodels may receive the same or different representations of the input values.

[0063] In step S6, the process can then either directly return the output value from the trained machine learning model or calculate an output value derived from the output of the machine learning model. In some embodiments, the process may perform one or more actions in response to the output value, for example, issue an alarm in response to a predicted error condition, or further execute one or more control actions to control a pump of another component of the system in response to the output of the machine learning model.

[0064] Figures 6 through 8 schematically show examples of machine learning models for use in embodiments of the methods disclosed herein.

[0065] In particular, FIG. 6 shows a machine learning model 120 that receives, as input 110, a representation of a pressure-stroke curve, i.e., a representation of a series of data points (P i , S i ), where i = 1,..., n, and each data point represents a pair of values (P i , S i , S i ) representing the pressure Pi and the corresponding displacement position S. The machine learning model can receive the series of data points as a column of doublets (P i , S i ), as a 2D image of a pressure-stroke diagram, or in another suitable form. The machine learning model outputs operation information 130 and is trained to calculate it. Examples of operation information include the classification of the operating conditions of a metering pump, particularly the classification of error conditions of the metering pump. Other examples include the values of operating parameters, particularly continuous value parameters such as discharge pressure, effective stroke length, and discharge flow rate.

[0066] FIG. 7 shows a pressure value P as input 140 iA machine learning model 150 is shown that is trained to receive a 1D representation and output an output value 130 indicative of operating information. The 1D representation can represent a time series of detected pressure values (or pressure values derived from detected values), or a series of detected or derived pressure values at respective positions during a stroke cycle. In some embodiments, the 1D representation may represent a single stroke cycle starting from a predetermined position during the stroke cycle, i.e., the 1D representation may be a representation of a pressure-stroke curve. In other embodiments, the 1D representation may represent another duration, e.g., more or less than a single stroke cycle, and alternatively or additionally, the 1D representation may represent a time window starting from an unknown position during the stroke cycle.

[0067] For this purpose, in some embodiments, the machine learning model 150 can be or include a regression network. Alternatively, the machine learning model can be or include a feed-forward network configured to receive a fixed-length one-dimensional input sequence representing pressure values sampled over a period of a predetermined length.

[0068] In either case, the machine learning model 150 outputs an output value 130 in response to receiving a time series as input. As described in connection with FIG. 6, the output value can indicate a classification of the operating state of the metering pump or indicate operating parameters of the metering pump.

[0069] Generally, it is desirable to provide a machine learning model that can be easily adapted to different types of pumps without the need to obtain a large training dataset for each individual type of pump.

[0070] For example, some metering pumps may not be able to output position information regarding the displacement of a displacement member. They may only be able to output pressure values or another type of indicator quantity. Thus, it is desirable to provide a method that can also calculate operating information for this type of metering pump.

[0071] In particular, the process can receive the time series of the detected pressure values without information on how the time series relates to the movement of the displacement member. In particular, the process may not know the relative phase of the time series with respect to the stroke cycle and / or the process may not know the duration of the stroke cycle with respect to the time series, i.e., how many cycles the time series covers.

[0072] Thus, FIG. 8 shows a machine learning model 850 that receives as input a time series representation 840 similar to that described in connection with FIG. 7. However, the time series 840 may cover more or fewer stroke cycles than a single stroke cycle, and the start of the time series with respect to the stroke cycle may also be unknown. The machine learning model 850 of FIG. 8 includes two sub-models 851 and 853, respectively. The first sub-model 851 is trained to receive the time series 840 and output a representation of a pressure-stroke curve 852, which can then serve as an input to the second sub-model 853. The representation of the pressure-stroke curve 852 covers a single stroke cycle and is in the form of a 1D representation of pressure values P i starting from a predetermined position during the stroke cycle. The 1D representation can then serve as an input to the second sub-model 853, for example, as the input 140 to the model of FIG. 7. Alternatively, the representation of the pressure-stroke curve 852 may be a representation of a series of data points (P i , S i )(i = 1,..., n), where each data point is a pair of values (P i and the corresponding displacement position S i representing pressure P i , S iIt represents . In any case, the second partial model 853 is similar to the model 120 in FIG. 6 or the model 150 in FIG. 7. For example, in response to receiving the representation of the pressure stroke diagram 852 as described in relation to FIG. 6 or FIG. 7, it may be trained to output an output value 130 indicating operation information. It should be understood that the representation of the pressure-stroke curve need not be a 1D representation of the pressure values corresponding to the cycle of the displacement member as shown in FIG. 8. Instead, it may be a 2D representation as shown in FIG. 7.

[0073] Therefore, the first partial model 851 only needs to be trained to convert the time series 840 of pressure values into the corresponding representation of the pressure stroke curve 852. Since the pressure changes periodically with the stroke cycle time, the conversion task corresponds to detecting the phase shift of the pressure time series with respect to the stroke cycle and / or the period of the stroke cycle with respect to the time series 840. The inventors have understood that a machine learning model can be trained to perform this task. Therefore, the second partial model 853 can be efficiently reused or at least adapted for different types of metering pumps by, for example, a suitable transfer learning process based on a relatively small number of additional training examples. In particular, the second partial model may even be used or adapted for use with a metering pump that provides position information regarding the displacement member. Similarly, the output of the first partial model 851 can be used as an input to a plurality of second partial models for detecting respective operating conditions or estimating respective operating parameters.

[0074] It will be appreciated that the first partial model may also be divided into a plurality of submodels, for example, a first submodel that receives a time series of pressure values and outputs a time series having a length corresponding to a single stroke cycle, preferably a fixed-length time series. The second partial model can receive the output of the first partial model and output a phase-shifted time series covering a single stroke cycle starting at a predetermined position of the displacement member, such as "bottom dead center". Depending on the type of pump, in particular depending on the position / cycle information available from a given pump, the first submodel, the second submodel or both submodels can be omitted.

[0075] It will be appreciated that in some embodiments, the machine learning model 850 can be implemented as a single model.

[0076] Figures 9 to 11 show examples of prediction results obtained by a method for determining the operating information of a metering pump as disclosed herein.

[0077] Figure 9 shows the results of a classification model trained to predict five different classes of error conditions of a metering pump from the representation of the pressure-stroke curve. The operating conditions included pumps operating normally and pumps operating under different types of malfunctions such as bubbles and cavitation. Figure 9 shows how the predicted classes correlate with the known actual operating conditions of the pump. Note that each point is intentionally shifted slightly from the classification value so as to be visually distinguishable. As can be seen from Figure 9, the classification results in a clear distinction between various operating conditions with only minor misclassifications, in other words, the trained machine learning model has been found to perform accurate and reliable classification.

[0078] Figures 10 and 11 show the correlations of two trained regression models, respectively. The model in Figure 10 was trained to predict the outlet pressure of a metering pump from a pressure-stroke curve, while the model in Figure 11 was trained to predict the effective stroke length of the metering pump. As can be seen from the correlations with the respective reference measurements, the predicted values by the methods disclosed herein provide accurate estimates of the operating parameters of the pump.

[0079] Embodiments of the methods described herein may be computer-implemented. In particular, embodiments of the methods may be implemented by hardware including several distinct elements and / or by a data processing system that is at least partially appropriately programmed. In apparatus claims enumerating several means, some of these means may be embodied by one and the same element, component or item of hardware. The mere fact that certain means are recited in mutually different dependent claims or are described in different embodiments does not indicate that combinations of these means cannot be used advantageously.

[0080] In summary, the various aspects disclosed herein may be summarized as follows. Embodiment 1: A method for determining operating information of a metering pump, the metering pump comprising a dosing chamber, a displacement member, and a drive motor for driving the displacement member, the method comprising: a) receiving a plurality of detected values of an indicator quantity indicative of the strength of the actuation of the displacement member at each position of the displacement member during operation of the metering pump; b) calculating the operating information from a machine learning model trained to output the operating information in response to receiving a plurality of input values derived from the detected values of the indicator quantity; comprising. Embodiment 2: The method according to Embodiment 1, wherein the indicator quantity includes the pressure in the dosing chamber and / or the torque of the drive motor. Embodiment 3: The method according to any one of the preceding embodiments, wherein the operation information includes classification of the operating conditions of the metering pump, particularly classification of error conditions of the metering pump. Embodiment 4: The method according to Embodiment 3, wherein the machine learning model includes a classification model trained to output an identifier of one of a plurality of individual classes. Embodiment 5: The method according to any one of the preceding embodiments, wherein the operation information includes values of operation parameters. Embodiment 6: The method according to Embodiment 5, wherein the machine learning model includes a regression model trained to output values of continuous value operation parameters. Embodiment 7: The method according to Embodiment 6, wherein the operation parameter indicates one or more of the operation parameters of discharge pressure, effective stroke length, and discharge flow rate. Embodiment 8: A method according to any one of the preceding embodiments, wherein the machine learning model is configured to receive a plurality of input values of an indicator quantity, each of the plurality of input values being associated with a respective position of a displacement member, and the machine learning model is configured to output the operation information in response to receiving the plurality of pairs of input data. Embodiment 9: The method according to Embodiment 8, · receiving position data indicating the monitored position of the displacement member during operation of the metering pump, or at least calculating position data from the detected values of the received indicator quantity, and · calculating a plurality of input values from the received detected values of the indicator quantity and from the received or calculated position data. A method comprising. Embodiment 10: The method according to any one of Embodiments 1 to 7, wherein the machine learning model is configured to receive a time series of detected values of the indicator quantity at each point in time and to output the operation information in response to receiving the time series of detection positions of the detected values of the indicator quantity. Embodiment 11: The method according to Embodiment 10, The machine learning model includes a first machine learning model and a second machine learning model. The first machine learning model is configured to calculate a plurality of input values of the indicator quantity based on the detected values of the received time series of the indicator quantity at each point in time during the operation of the metering pump. Each input value indicates the value of the indicator quantity at each position of the displacement member, and the second machine learning model is configured to output operation information in response to receiving the calculated plurality of input values. Embodiment 12: A computer-implemented method for creating a trained machine learning model for use in the method according to Embodiment 1, wherein the training method comprises: a) obtaining a set of training data items, each training data item including a plurality of input values and a corresponding target output, the plurality of input values indicating an indicator quantity indicating the intensity of the operation of the displacement member of the metering pump at each position of the displacement member during the operation of the metering pump, and the corresponding target output indicating operation information observable during the operation of the metering pump; b) training a machine learning model from the obtained set of training data to output operation information in response to receiving the plurality of input values; and including. Embodiment 13: A data processing system configured to execute the steps of the method defined in any one of Embodiments 1 to 12. Embodiment 14: A metering pump comprising a dosing chamber, a displacement member, a drive motor for driving the displacement member, and the data processing system defined in Embodiment 13. Embodiment 15: A system including the metering pump and the data processing system defined in Embodiment 13, the metering pump comprising a dosing chamber, a displacement member, and a drive motor for driving the displacement member. Embodiment 16: The system according to Embodiment 15, wherein the data processing system is separate from the metering pump and includes an interface for receiving a plurality of detected values of an indicator quantity indicating the intensity of the operation of the displacement member at respective positions of the displacement member during operation of the metering pump. Embodiment 17: The system according to Embodiment 15, wherein the metering pump further includes a data processing system. Embodiment 18: A computer program including computer program code configured to cause the data processing system to execute the steps of the method according to any one of Embodiments 1 to 12 when executed by the data processing system.

[0081] As used herein, the term "comprising" is to be interpreted as specifying the presence of the stated features, elements, steps, or components, but not precluding the presence or addition of one or more other features, elements, steps, components, or groups thereof.

Claims

1. A method for determining operating information of a metering pump, the method comprising: the metering pump comprising a dosing chamber, a displacement member, and a drive motor for driving the displacement member, the method comprising: a) receiving a plurality of detected values of an indicator quantity indicative of the strength of actuation of the displacement member at each position of the displacement member during operation of the metering pump; b) calculating the operating information from a machine learning model trained to output the operating information in response to receiving a plurality of input values derived from the detected values of the indicator quantity; A method as described above.

2. The indicator quantity includes the pressure in the dosing chamber and / or the torque of the drive motor, The method according to claim 1.

3. The operating information includes a classification of the operating conditions of the metering pump, in particular a classification of error conditions of the metering pump, The method according to any one of the preceding claims.

4. The machine learning model includes a classification model trained to output an identifier of one of a plurality of individual classes, The method according to claim 3.

5. The operating information includes values of operating parameters, The method according to any one of the preceding claims.

6. The machine learning model includes a regression model trained to output values of continuous value operating parameters, The method according to claim 5.

7. The operating parameters indicate one or more of the operating parameters of discharge pressure, effective stroke length, and discharge flow rate, The method according to claim 5 or 6.

8. The machine learning model is configured to receive the plurality of input values of the indicator quantity, each of the plurality of input values being associated with a respective position of the displacement member, the machine learning model being configured to output the operating information in response to receiving at least the plurality of input values, The method according to any one of the preceding claims.

9. receiving position data indicative of the monitored position of the displacement member during operation of the metering pump, or calculating position data from at least the detected values of the indicator quantity received, and calculating the plurality of input values from the received detected values of the indicator quantity and from the received or calculated position data, The method according to claim 8.

10. The machine learning model is configured to receive a plurality of pairs of input data, each pair of input data including the position of the displacement member and the corresponding value of the indicator amount at the position, the machine learning model being configured to output the operation information in response to receiving the plurality of pairs of input data, The method according to claim 8 or 9.

11. The machine learning model is configured to receive a time series of detected values of the indicator amount at each point in time, and to output the operation information in response to receiving the time series of detected values of the indicator amount, The method according to any one of claims 1 to 9.

12. The machine learning model includes a first machine learning model and a second machine learning model, the first machine learning model being configured to calculate the plurality of input values of the indicator amount based on a time series of received detected values of the indicator amount at each point in time during operation of the metering pump, each input value indicating the value of the indicator amount at each position of the displacement member, the second machine learning model being configured to output the operation information in response to receiving the calculated plurality of input values, The method according to claim 11.

13. A computer-implemented method for creating the trained machine learning model for use in the method according to any one of claims 1 to 12, the training method comprising: a) obtaining a set of a plurality of training data items, each of the plurality of training data items including a plurality of input values and a corresponding target output, the plurality of input values indicating the indicator amount indicating the intensity of the operation of the displacement member of the metering pump at each position of the displacement member during operation of the metering pump, and the corresponding target output indicating the operation information observable during the operation of the metering pump; and b) training the machine learning model from the obtained set of training data to output the operation information in response to receiving the plurality of input values. Computer-implemented method.

14. A data processing system configured to execute the steps of the method according to any one of claims 1 to 13.

15. A dosing chamber, a displacement member, a drive motor for driving the displacement member, and the data processing system according to claim 14, A metering pump comprising the same.

16. A system comprising a metering pump and the data processing system according to claim 15, wherein the metering pump comprises a dosing chamber, the displacement member, and the drive motor for driving the displacement member, A system comprising the same.

17. The data processing system is separate from the metering pump and comprises an interface for receiving a plurality of detected values of the indicator quantity indicating the strength of the operation of the displacement member at respective positions of the displacement member during operation of the metering pump, The system according to claim 16.

18. The metering pump further comprises a data processing system, The system according to claim 16.

19. A computer program comprising computer program code configured to cause the data processing system to execute the steps of the method according to any one of claims 1 to 13 when executed by the data processing system.

Images