Method for detecting an operating state of a centrifugal pump and centrifugal pump

The method of Fourier transforming and analyzing centrifugal pump signals using machine learning detects blockages and maintenance needs, addressing the reliability issues of current methods and preventing pump damage.

EP4461964A1Pending Publication Date: 2024-11-13WILO SE

Patent Information

Application Number
EP2024165437
Authority / Receiving Office
EP · EP
Patent Type
Applications
Current Assignee / Owner
Priority Date
2023-05-10
Filing Date
2024-03-22
Publication Date
2024-11-13

AI Technical Summary

Technical Problem

Current methods for detecting operating states and blockages in centrifugal pumps are not reliable or timely, leading to potential damage from solids content, which can cause clogging and inefficiency.

Method used

A method involving Fourier transformation of acceleration, pressure, and motor current signals to obtain frequency spectra, followed by analysis using a machine-learning trained image analysis algorithm to recognize spectral patterns, allowing for early detection of blockages and maintenance needs.

Benefits of technology

This approach enables reliable and reproducible detection of operating states, including blockages and solids content, allowing for timely intervention to prevent damage and improve pump efficiency.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure IMGAF001_ABST
    Figure IMGAF001_ABST
Patent Text Reader

Abstract

The invention relates to a method for detecting an operating state of a centrifugal pump (1) comprising a pump housing (2) with a shaft (4) arranged in the pump housing (2), an impeller (6) mounted on the shaft (4) and a motor (5) driving the shaft (4), comprising the steps of: during pumping of liquid with a possible solid content with the centrifugal pump (1), acquiring a temporal acceleration, pressure and / or motor current signal of the centrifugal pump (1), Fourier transforming the acceleration, pressure and / or motor current signal to obtain a frequency spectrum and / or a spectrogram, and analyzing the frequency spectrum and / or the spectrogram with regard to a spectral pattern to detect the operating state.
Need to check novelty before this filing date? Find Prior Art

Description

Technical field

[0001] The invention relates to a method for detecting the operating state of a centrifugal pump comprising a pump housing with a shaft arranged in the pump housing, an impeller mounted on the shaft, and a motor driving the shaft, comprising the step of detecting a temporal acceleration, pressure, and / or motor current signal of the centrifugal pump while pumping a liquid containing a possible solid content. The invention further relates to a centrifugal pump comprising a pump housing with a shaft arranged in the pump housing, an impeller mounted on the shaft, a motor driving the shaft, and a control unit, wherein the control unit is configured to detect a temporal acceleration, pressure, and / or motor current signal of the centrifugal pump while pumping a liquid containing a possible solid content. Background of the invention

[0002] Centrifugal pumps are known from the prior art and are used to pump a liquid by means of the rotary motion of an impeller. The liquid to be pumped enters a pump chamber of the centrifugal pump through a suction opening, is captured by the rotating impeller and subsequently conveyed through a discharge port.

[0003] Solid particles contained in the liquid can settle in the area of ​​the impeller and on the inside of a pump housing, negatively affecting the hydraulic and / or mechanical efficiency of the centrifugal pump, potentially leading to clogging and failure of the centrifugal pump.

[0004] Although several methods for detecting an operating condition and, in particular, for resolving blockages in centrifugal pumps are known from the prior art, current practice shows that the known methods are not ideal for protecting a centrifugal pump from damage due to such a blockage that may not be detected in time in a safe and reliable manner, and in particular for reliably detecting a corresponding operating condition of the centrifugal pump. Description of the invention

[0005] Starting from this situation, it is an object of the present invention to provide a method for detecting an operating state of a centrifugal pump and a corresponding centrifugal pump which, compared to solutions known from the prior art, can detect the operating state, and in particular a blockage, more reliably and quickly and can initiate measures to resolve the blockage more quickly, for example.

[0006] The object of the invention is achieved by the features of the independent claims. Advantageous embodiments are specified in the dependent claims.

[0007] Accordingly, the problem is solved by a method for detecting an operating state of a centrifugal pump comprising a pump housing with a shaft arranged in the pump housing, an impeller sitting on the shaft and a motor driving the shaft, with the following steps: During the pumping of liquid with a possible solid content using the centrifugal pump, a temporal acceleration, pressure and / or motor current signal of the centrifugal pump is acquired, the acceleration, pressure and / or motor current signal is Fourier-transformed to obtain a frequency spectrum and / or a spectrogram, wherein in particular frequencies of the frequency spectrum and / or the spectrogram are normalized, especially by means of a rotational speed of the centrifugal pump, and the frequency spectrum and / or the spectrogram is analyzed, especially by means of an image analysis algorithm trained by machine learning, with regard to at least one spectral pattern to recognize the operating state.

[0008] A key aspect of the proposed teaching is that, particularly with the help of a machine-trained image analysis algorithm, the operating state of the centrifugal pump can be reliably and reproducibly derived from the spectral pattern obtained in the Fourier-transformed acceleration, pressure, and / or motor current signal. This surprising technical effect allows, for example, experimental verification of whether the centrifugal pump is clogged, whether a maintenance cycle is necessary, the level of solids in the fluid, and whether an abnormal operating condition or bearing failure exists. By determining the operating state in this way, it is possible, for instance, to react earlier to an impending blockage, thus reducing or preventing damage to the centrifugal pump caused by solids.The proposed method offers a completely novel approach compared to prior art methods for detecting the operating state of a centrifugal pump, for example, to signal it to a downstream control system and / or to initiate appropriate measures based on the detected operating state. Instead of using an image analysis algorithm trained by machine learning, the analysis of the frequency spectrum and / or the spectrogram can be performed using a thresholding method with respect to at least one spectral pattern for detecting the operating state.

[0009] A centrifugal pump is generally defined as a turbomachine that utilizes rotary motion and dynamic forces to pump predominantly liquids. Centrifugal pumps are preferably designed as wastewater pumps. In addition to the tangential acceleration of the liquid in radial flow, centrifugal force is used for pumping, so such pumps are also referred to as centrifugal pumps. Centrifugal pumps are preferably used in the hydraulic system of a building or other applications.

[0010] In normal operation of the centrifugal pump, the motor housing can be located above the pump housing. The impeller, driven by the motor via a shaft, is located within the motor housing and is used to pump the fluid. The motor housing can be fixedly connected to the pump housing and / or be a single unit. The centrifugal pump and the motor can each have their own shaft, which can be connected via a coupling. Preferably, the shaft projects from the motor housing into the pump housing on one drive side and / or is fixedly connected to the shaft on the drive side of the impeller. Accordingly, an inlet for the fluid to be pumped can preferably be located below or at the bottom of the pump housing.

[0011] The fluid preferably comprises water or another liquid medium such as wastewater. The fluid may contain solids such as impurities or waste of any kind, in particular feces, sediments, dirt, sand, or even small pieces of wood, brush, textiles, or rags, or the like. Preferably, the motor housing and / or the pump housing is made of metal, in particular cast iron or stainless steel, ceramic, and / or plastic. The operating condition includes, in particular, a normal operating condition, i.e., regular operation of the centrifugal pump, or an abnormal operating condition, in particular a malfunction and / or a blockage, and / or information, for example, about the amount of solids, a required maintenance cycle, or a bearing failure.

[0012] The phrase 'during the pumping of liquid' refers specifically to the regular operation of the centrifugal pump, namely when the centrifugal pump is pumping a fluid that may contain the aforementioned solids. Accordingly, by operating the motor in the first and / or second direction of rotation, the impeller can be driven forward and / or backward, particularly for pumping the liquid or fresh water as described below. 'Liquid with possible solids' means, in this context, that during regular operation only liquid is pumped, but the liquid may contain the generally undesirable solids, for example, partially or even predominantly.

[0013] A temporal acceleration, pressure, and / or motor current signal is understood to be, in particular, a signal characterizing the centrifugal pump during operation, for example, the electrical current consumed by the centrifugal pump and / or the electrical power consumed by the motor while pumping the fluid, a rotational speed, and / or a torque of the motor. Fourier transforming the acceleration, pressure, and / or motor current signal, also referred to as the signal in this application, results in translational invariance of the signal in both the time and frequency directions. A Fourier transform, more precisely a continuous Fourier transform, is generally understood to be a method from the field of Fourier analysis used to decompose aperiodic signals into a continuous spectrum.The function that describes this spectrum is also called the Fourier transform or spectral function, which is obtained via an integral transformation.

[0014] The frequency spectrum of a signal, usually simply called the spectrum, shows the signal's composition from different frequencies. A spectrogram is generally understood as a graphical representation of the signal's frequency spectrum over time. Machine learning is preferably performed using a neural network, especially an artificial neural network. The image analysis algorithm can be run on a computer using a processor and / or by a controller as described below. The term "machine learning" should preferably also be understood to include, for example, the recognition of equidistant peaks in the frequency spectrum.

[0015] According to a preferred embodiment, the spectral pattern for detecting the operating state is characterized by exceeding an acceleration, pressure, and / or motor current threshold, in particular an integer and / or half-order of the signal. Order or order analysis is generally understood as an analysis of the signal in which the signal's energy content is plotted against the order. The threshold can, in principle, be defined arbitrarily; for example, in an abnormal operating state, it might be 10, 20, or 30% higher than in a regular operating state, particularly as a deviation from or relation to a mean value of all other regions in the frequency spectrum and / or spectrogram.Therefore, if the Fourier-transformed signal, in particular of the integer and / or half order of the signal, exceeds the threshold, a deviation of the operating state from the regular operating state can be inferred, and the deviation can be signaled and / or rectified by clearing a blockage, for example.

[0016] According to a further preferred embodiment, the spectral pattern for detecting the operating state is characterized by exceeding a predefined time period, in particular a moving average of the acceleration, pressure, and / or motor current threshold. In a visual representation, the spectral pattern can be characterized as horizontal "rungs" of a "ladder," i.e., when, in particular, the integer and / or half order of the signal exceeds the threshold for the predefined time period. The threshold can be, for example, 1, 2, 5, 10, 20, 30, or 60 seconds.This so-called spectral ladder, consisting in particular of multiple rungs, represented in the spectrogram especially as horizontal lines corresponding to orders and semi-orders, and usually two arms, represented in the spectrogram especially as vertical lines, for example corresponding to the beginning and end of a blockage, is visually apparent primarily as a distinct color difference compared to the local environment. The color difference arises primarily from changing amplitude values ​​along a third axis of the spectrogram, specifically orthogonal to the image plane, which can be represented by a color chart. Depending on the choice of color chart and the color representations of the minimum and maximum amplitude values, a corresponding color difference can result.

[0017] According to another preferred further development, the method comprises the step of analyzing the frequency spectrum and / or the spectrogram in a range between the 3rd and 9th order, in particular between the 3rd and 5th, 6th, 7th or 8th order of the signal.

[0018] In another preferred configuration, machine learning is performed using a neural network, specifically a convolutional neural network (CNN). A convolutional neural network is generally understood to be an artificial neural network, with classic convolutional neural networks consisting of one or more convolutional layers, each followed by a pooling layer and one or more fully connected layers. The convolutional layer receives an input, often in the form of a one-, two-, or three-dimensional matrix, such as pixels in a grayscale or color image.

[0019] According to another preferred further development, the method includes the step of analyzing the frequency spectrum and / or the spectrogram using the image analysis algorithm trained by machine learning with regard to several and / or different spectral patterns to detect the operating state.

[0020] In particular, by using multiple CNN color channels, especially multiple parallel CNN input channels, for additional images, different physical meanings can be derived from the frequency spectrum and / or the spectrogram of different time signals. For example, multiple spectral patterns can be created to detect the operating state. This allows the operating state to be captured in significantly greater detail.

[0021] According to a further preferred embodiment, the procedure comprises the following steps: Analyzing the spectral pattern with regard to a blockage as an operating condition of the centrifugal pump; upon detection of the blockage as an operating condition, resolving the blockage by operating the motor in at least one second direction of rotation opposite to a first direction of rotation for pumping the liquid.

[0022] To resolve the blockage, so-called unblocking routines known from the prior art can generally be used, which usually involve a time-limited and / or repetitive, in particular interval-repeated and / or alternating operation of the motor in the first and / or second direction of rotation, especially in forward and / or reverse directions.

[0023] According to another preferred training method, the procedure includes the following steps: Upon detection of the blockage as an operating condition, the spectral pattern is analyzed with regard to a classification of the blockage of the centrifugal pump, and the blockage is resolved by operating the motor depending on the classification of the blockage.

[0024] According to this design, blockages can be resolved much more effectively compared to the prior art, since the method for clearing the blockage, in particular the initiated unblocking routine, is selected based on the previously determined classification of the blockage. This is because the spectral pattern reveals the type of blockage, as has been experimentally demonstrated. For example, a clearer or more distinct pattern indicates that larger or heavier contaminants are trapped in the impeller than in a less distinct pattern. Similarly, the spectral pattern allows for the localization of a blockage.To clear the blockage, a suitable unblocking routine can be selected, tailored in one case to larger or harder contaminants and in another case to smaller and softer contaminants, or tailored to the location of the blockage. Such unblocking routines can be characterized by lower / higher rotational speeds, lower / higher accelerations, and / or fewer / more intervals.

[0025] According to a further preferred embodiment, the method includes the step of: capturing the signal by means of an acceleration and / or a vibration sensor, in particular by means of a 3-axis acceleration and / or a vibration sensor, by means of a pressure sensor and / or by means of a current sensor.

[0026] Such sensors can, in principle, be designed as known from the prior art, for example, as a uniaxial and / or triaxial accelerometer. The accelerometer and / or vibration sensor is preferably located in contact with and / or directly adjacent to the pump housing, shaft, impeller, motor, and / or a motor housing. The pressure sensor can be located on the suction side and / or pressure side. The current sensor is preferably arranged in the current path of the motor, particularly on a connecting cable of the motor.

[0027] According to another preferred further development, the procedure includes the step: obtaining the frequency spectrum by means of fast Fourier transform and / or the spectrogram by means of short-time Fourier transform.

[0028] The term "fast Fourier transform" (FFT) generally refers to an algorithm for efficiently calculating the discrete Fourier transform (DFT), which decomposes a discrete-time signal into its frequency components for analysis. The short-time Fourier transform (STFT) can be used to represent the temporal changes in the signal's frequency spectrum. While the Fourier transform does not provide information about the temporal evolution of the spectrum, the STFT is also suitable for signals whose frequency characteristics change over time.

[0029] According to a further preferred embodiment, the method includes the step of training the image analysis algorithm using a large number of individual images with at least partially different operating states to obtain a respective vector with probabilities of a regular operating state and an abnormal operating state, in particular of non-blockage and blockage.

[0030] Advantageously, such training is carried out using a large number of individual images, particularly several hundred, for example 800, 1000, or 1500. This training can be followed by a test phase with several more, particularly hundreds, of individual images, or the test phase can take place during the training. The individual images preferably contain corresponding spectral patterns of the regular operating state as well as the abnormal operating state, for example, when the centrifugal pump is unblocked and when it is blocked. The patterns used for training may not always be perceptible to the human eye.

[0031] According to another preferred further training, the procedure includes the step: training the image analysis algorithm when pumping clear water as a liquid and liquid containing a defined solids content.

[0032] This allows the spectral pattern of the centrifugal pump to be determined under normal operating conditions, enabling the classification of deviations as abnormal operating conditions. For this purpose, the solids content is preferably defined in such a way that a blockage constitutes an abnormal operating condition. The solids content can also be defined by one or more particles, particularly of a predetermined size, shape, etc., and / or per unit of time.

[0033] There are several ways to trigger the proposed procedure, for example, based on a temporal event and / or the result of another procedure. However, in a particularly preferred embodiment, the procedure comprises the following steps: During the pumping of liquid with a possible solid content, detecting a slope and / or curvature of a time course of the centrifugal pump signal, and if the slope and / or curvature of the time course of the signal exceeds a blockage threshold, Fourier transforming the signal.

[0034] By using such gradient information and / or the curvature of the signal's time course, a quasi-estimated glimpse into the near future can be obtained. In other words, the slope and / or curvature of the signal's time course allows for an estimation of how the signal will change in the immediate future, for example, in the next few milliseconds. Put another way, the proposed enhancement provides a kind of trend analysis for the signal, enabling the detection of a trend toward a change in the centrifugal pump's operating state within just a few milliseconds. This allows for a more detailed understanding of the operating state using the Fourier transform. The trend identified in this way allows for earlier intervention, such as reacting to an impending blockage, thus reducing or preventing damage to the wastewater pump caused by contaminants.

[0035] The slope is defined in particular as a first time derivative of the signal's time course, especially of the electrical operating parameters characterizing the rotating motor, for example, the electrical current and / or power consumed by the motor during conveying, speed, or torque, and in particular the rate of change of the electrical operating parameters characterizing the rotating motor. The first derivative is commonly referred to as d f / d t described.

[0036] Curvature is understood to mean, in particular, the time derivative of the slope of the signal's time course, or the twofold derivative of the signal's time course, especially of the electrical operating parameters characterizing the rotating motor, with respect to time. Electrical operating parameters characterizing the rotating motor include, for example, the electrical current and / or power consumed by the motor during fluid transport, or the motor's torque.

[0037] Within the scope of the invention, a temporal profile is understood to be, in particular, a temporally ordered sequence of the signal, especially signal values, particularly operating values ​​characterizing the motor, such as power, torque, speed, or current, and in particular a time series. The temporal profile can be averaged, in particular quadratically averaged, and / or smoothed. The temporal profile can comprise a continuous and / or discrete sequence of the signal. Particularly preferably, the slope and / or curvature of the temporal profile comprises an approximate derivative of discrete time signals by difference quotients of a first and / or second order of the operating value.

[0038] According to another preferred embodiment, the method includes the step of capturing the signal with a sampling frequency of ≥ 10 kHz and ≤ 60 kHz.

[0039] Preferably, the sampling frequency is 20, 30, 40, or 50 kHz. Lower frequency limits such as 1, 2, 5, or 8 kHz are also conceivable.

[0040] The object of the invention is further solved by a centrifugal pump comprising a pump housing with a shaft arranged in the pump housing, an impeller sitting on the shaft, a motor driving the shaft and a control system, wherein The control system is configured to acquire a temporal acceleration, pressure and / or motor current signal of the centrifugal pump during the pumping of liquid with a possible solid content, to Fourier transform the signal to obtain a frequency spectrum and / or a spectrogram, wherein in particular frequencies of the frequency spectrum and / or the spectrogram are normalized by means of a rotational speed of the centrifugal pump, and to analyze the frequency spectrum and / or the spectrogram, in particular by means of an image analysis algorithm trained by machine learning, with regard to at least one spectral pattern to recognize an operating state of the centrifugal pump.

[0041] The control unit preferably comprises a microprocessor or the like, by which the signal can be acquired, the signal Fourier-transformed, and / or the frequency spectrum and / or the spectrogram analyzed. The control unit can be an integral part of the centrifugal pump, provided on the centrifugal pump, for example, on a housing of the centrifugal pump, arranged remotely from the centrifugal pump, and / or be part of a "cloud," particularly one remotely from the centrifugal pump. For acquiring the signal, a sensor, for example, an acceleration, pressure, and / or current sensor, connected directly or indirectly to the motor, shaft, impeller, and / or the centrifugal pump, particularly to a housing, motor housing, and / or pump housing, can be provided, and / or monitor the motor, shaft, impeller, and / or centrifugal pump.

[0042] Further features and advantages of the centrifugal pump become apparent to the expert by analogy to the procedure described above. Brief description of the drawings

[0043] The invention is explained in more detail below with reference to the accompanying drawings and by way of preferred embodiments.

[0044] The drawings show Fig. 1 a schematic view of a centrifugal pump for carrying out the proposed method according to a preferred embodiment of the invention, Fig. 2 a flowchart for carrying out the proposed method according to the preferred embodiment of the invention, Fig. 3a shows a spectrogram of with a rotational speed of the centrifugal pump of the Fig. 1 normalized frequencies of a time-dependent signal according to the preferred embodiment of the invention, Fig. 3b shows a section of the spectrogram of the Fig. 3aAccording to the preferred embodiment of the invention, Fig. 4 shows the spectrogram of at a rotational speed of the centrifugal pump of Fig. 1 normalized frequencies of the time signal according to the preferred embodiment of the invention in a further view, Fig. 5 shows above a motor of the centrifugal pump of the Fig. 1 Fig. 6 shows the recorded motor current, in the middle the spectrogram of an acceleration signal, and below a probability of clogging of the centrifugal pump according to the preferred embodiment of the invention. Fig. 6 shows a section of a power / time diagram when carrying out the proposed method according to the preferred embodiment of the invention, and Fig. 7 shows a power / angular frequency diagram when carrying out the proposed method according to the preferred embodiment of the invention. Detailed description of the implementation examples

[0045] Fig. 1Figure 1 shows a schematic view of a centrifugal pump 1 for carrying out a method described below for determining an operating state and in particular for resolving a blockage of the centrifugal pump 1 according to a preferred embodiment of the invention.

[0046] The centrifugal pump 1 has a conventional pump housing 2 with a suction opening 3, located at the bottom of the pump housing 2 in the figure, serving as the inlet. A shaft 4 is provided in the pump housing 2, extending vertically in the drawing. The shaft 4 is driven by a motor 5, shown only partially, which is arranged opposite the suction opening 3. An impeller 6 is provided facing the suction opening 3 and is driven by the motor 5 via the shaft 4. Furthermore, the centrifugal pump 1 has a microprocessor-based control unit 7, in Fig. 1 only hinted at.

[0047] Finally, the centrifugal pump 1 has a sensor 8 for detecting a temporal acceleration, pressure, and / or motor current signal from the centrifugal pump 1. The sensor 8 is configured accordingly as an acceleration and / or vibration sensor, in particular as a 3-axis acceleration and / or vibration sensor, as a pressure sensor, and / or as a current sensor. In the case of an acceleration and / or vibration sensor configuration, the sensor 8 is provided, in particular, in contact with, directly adjacent to, and / or rigidly mounted, for example, on a metal plate, on the motor 5, associated with the shaft 4, on the pump housing 2, on a motor housing of the motor 5, and / or associated with the impeller 6. In the case of a pressure sensor, the sensor 8 is provided on the pressure side and / or suction side, in particular at the suction opening 3. In the case of a current sensor, the sensor 8 measures a motor current drawn by the motor 5.

[0048] The proposed method for detecting the operating state of centrifugal pump 1, in a first step, records the acceleration, pressure, and / or motor current signal of centrifugal pump 1 over time during the pumping of liquid containing a possible solids content. For this purpose, the signal is recorded discontinuously over time at a sampling frequency of ≥ 10 kHz and ≤ 60 kHz, particularly 30 kHz. The sampling frequency can also be specified lower or higher.

[0049] The acceleration, pressure, and / or motor current signal thus recorded over time is analyzed in a second step 200 using Fourier transformation to generate a frequency spectrum and / or a spectrogram, represented in Fig. 3a, to obtain, where frequencies of the frequency spectrum and / or the spectrogram are normalized by a rotational speed of centrifugal pump 1. Specifically, the frequency spectrum is obtained using a fast Fourier transform and / or the spectrogram using a short-time Fourier transform.

[0050] In a subsequent third step, the frequency spectrum and / or the spectrogram is analyzed using a machine learning-trained image analysis algorithm to identify a spectral pattern for determining the operating state. This is done as illustrated in the following example. Fig. 3b , the spectrogram is considered in a range between the 3rd and 9th order of the signal. In the case of a blockage of the centrifugal pump 1, for example because the solid content clogs or jams the impeller 6 or represents an additional frictional component, a result is obtained in Fig. 3bThe spectral "ladder" shown, with integer and half orders of the signal, is marked by arrows as horizontal "ladder rungs".

[0051] This spectral "ladder" as a spectral pattern represents, on the one hand, a relative exceedance of an acceleration, pressure, and / or motor current threshold of the integer and / or half order of the signal, which in Fig. 3a and Fig. 3b The signal's visibility is represented as a color difference or as the visibility of the Fourier-transformed signal along the frequency axis. Furthermore, a "ladder rung" is formed as a horizontal line of this signal due to exceeding a predefined time duration and / or the acceleration, pressure, and / or motor current threshold, or as the visibility of the Fourier-transformed signal along the time axis. The resulting spectral pattern characterizes the operating state of centrifugal pump 1 as a blockage.

[0052] In other words, to identify the conductors in the frequency spectrum, a preliminary step can be to convert any color-coded spectrogram into a black-and-white spectrogram, or in fact, such a conversion may already exist. This involves defining a black-and-white scale according to the following scheme: pure black, i.e., relative color value equal to 0, corresponding to the lowest absolute value in the spectrogram, and pure white, i.e., relative color value equal to 1, corresponding to the highest absolute value in the spectrogram. Additionally, a time window for analysis, which can be shifted over time, is defined, for example, 1, 2, or 3 seconds; that is, the analysis window is shorter than the duration of a single frame.

[0053] To detect whether a conductor is present in the frequency spectrum, we in Fig. 4As indicated, all whole and half orders between the 3rd and 9th, including the two limits, are considered within the analysis window; that is, orders 3 / 3.5 / 4 / 4.5 / 5 ... 8 / 8.5 / 9, hereinafter referred to as Max lines. Within the analysis window, the respective lines that lie a quarter order below and above all Max lines are considered; hereinafter referred to as Min lines. Specifically, in the case of order 3, the lines of orders 2.75 and 3.25; in the case of order 3.5, the lines of orders 3.25 and 3.75, and so on, up to the case of order 9, the lines of orders 8.75 and 9.25.

[0054] Around each minimum and maximum line, individual areas or strips are defined with a width equal to the analysis window and half-heights of, for example, + / - 0.02 orders above and below the respective line. This means each area has a total height H of 0.04 orders and is centered around the respective line. An average relative color value, referred to hereafter as the area value, is calculated for each area. Finally, the area values ​​of all maximum lines and the two immediately adjacent minimum lines are compared to determine whether the area value of the maximum line significantly exceeds the area values ​​of the two adjacent minimum lines. A value of, for example, 0.1 is used as the threshold for the difference in area values, expressed as a relative color value. In other words, a zigzag pattern is sought, which serves as a criterion for a ladder structure.The above can be applied analogously to a frequency spectrum.

[0055] In preparatory steps for executing the previously described procedure for detecting the operating state of centrifugal pump 1, the image analysis algorithm is first trained using a large number of individual images. This is shown in Fig. 5 The top graph shows the motor current recorded by motor 5, the middle graph shows the spectrogram, and the bottom graph shows the probability of a blockage. For example, a blockage was simulated by training with corresponding individual images during the period between approximately 550 and 650 seconds, so that the spectral "ladder" formed in the middle spectrogram, and consequently, the probability of a blockage at the bottom is 100%.

[0056] Machine learning is performed using a Convolutional Neural Network (CNN), i.e., an artificial neural network. The individual images, matrix objects with a resolution of NxN pixels, for example, 60x60 or 120x120 pixels, contain different operating states as input for the CNN, such as regular and abnormal operating states, specifically non-blockage and blockage, or normal bearing condition and bearing damage of centrifugal pump 1. An output of the CNN is a vector with two or more entries corresponding to the probabilities for the regular and abnormal operating states, specifically non-blockage and blockage, or normal bearing condition and bearing damage, with a total probability of 1.

[0057] The spectral pattern can be derived from standardized sections of the calculated spectrograms, in particular by adjusting and standardizing the image resolution to NxN pixels, for example, 60x60 or 120x120 pixels, and calculating numerous images, or as sections from spectrograms with and without features of a blockage or abnormal operating condition for training, any intermediate tests, and the subsequent final validation. For example, the image analysis algorithm of the CNN can be trained when pumping clear water as a liquid, liquid containing a defined maximum solids content that still allows normal operation of the centrifugal pump, and when centrifugal pump 1 is blocked.

[0058] Continuing the previously described procedure, in a further step 400 the spectral pattern can be analyzed to determine whether the centrifugal pump 1 is in a blocked operating condition. For this purpose, the frequency spectrum and / or the spectrogram can be examined using the machine learning-trained image analysis algorithm for multiple and / or different spectral patterns to identify a specific type of operating condition. By using multiple CNN input channels, different physical meanings can be determined, for example, the type of blockage can be detected. In addition to a general operating condition, the amount of solids, a required maintenance cycle, or a bearing defect can also be identified.

[0059] Therefore, upon detection of a blocked operating condition in a subsequent step 500, the spectral pattern can be analyzed to classify the blockage of the centrifugal pump. Subsequently, in a further step 600, the blocked operating condition can be resolved by operating the motor 5, depending on the classification of the blocked operating condition. Such resolution is achieved, for example, by operating the motor 5 in at least one second direction of rotation opposite to the first direction of rotation for pumping the liquid.

[0060] The above-described procedure can be initiated in particular if, during the pumping of liquid with the possible solid content, a slope and / or a curvature of a time course of the signal of centrifugal pump 1, i.e. the first and / or the second time derivative of the signal, exceeds a blockage threshold, so that in this case the Fourier transform of the signal takes place.

[0061] In preparatory steps for execution, the motor 5 can be repeatedly operated in forward and / or reverse directions while pumping clear water, i.e., in a second direction of rotation opposite to the first direction of rotation and / or in the first direction of rotation, and can also be accelerated in forward and / or reverse directions. During this operation of the motor 5 by means of a speed-time function specified by the controller 7, the sensor 8 records several discrete electrical clear water operating parameters characterizing the rotating motor as a function of time.

[0062] The controller 7 then determines the slope and / or curvature of the time course of the clear water operating values ​​thus obtained, i.e., the first and second time derivatives of the clear water operating values ​​thus obtained. Since the motor 5 was operated at different speeds, accelerations, and directions of rotation, characteristic resistance curves for the operation of the centrifugal pump 1 with clear water can be determined and stored in a memory of the controller 7. These resistance curves represent the basic operating behavior of the centrifugal pump 1 with clear water free of impurities.

[0063] When pumping liquid containing solids, i.e., water with various types of impurities, centrifugal pump 1 regularly requires more electrical energy than when pumping clear water. Accordingly, the slope and / or curvature of the electrical operating values ​​for wastewater pumping, as measured during pumping liquid containing solids, is higher than the corresponding slope and / or curvature of the operating values ​​for clear water pumping. These wastewater operating values ​​can, in turn, be measured by sensor 8.

[0064] If the centrifugal pump 1 becomes clogged, for example, if contamination reduces the speed of the impeller 6 or even jams the impeller 6, causing it to stop, the wastewater operating parameters change, namely the slope and / or the curvature of the time-dependent curve of the wastewater operating parameters. If the slope and / or the curvature of the time-dependent curve of the wastewater operating parameters exceeds a predefined blockage threshold, the controller 7 detects the blockage of the centrifugal pump 1.

[0065] The blockage threshold is conveniently calculated as the maximum of the slope and / or curvature of the clear water operating parameters, increased by a factor of 1.3. For example, it is 1.3 times the maximum slope and / or curvature of the clear water operating parameters. If the centrifugal pump 1 consumes 1000 W when pumping clear water, its regular power consumption when pumping liquid containing solids is, for example, 1050 W. However, as soon as the wastewater pump 1 consumes 1300 W, the controller 7 detects the blockage or a change in the operating state and begins the Fourier transform of the signal.

[0066] Examples include: Fig. 6 an excerpt from a performance / time diagram and in Fig. 7 A power / angular frequency diagram is shown during the execution of the proposed procedure. Fig. 6Figure 1 illustrates how the slope is derived from the power-time function and / or the power-angular frequency function as a wastewater operating value. From this slope, the controller 7 can detect when the centrifugal pump 1 becomes clogged and a clogging threshold is exceeded. Fig. 7 This illustrates various power / rotational frequency curves. The flatter solid power / rotational speed curve represents the acceleration behavior of centrifugal pump 1 without clogging when pumping clear water, with the respective lower and upper clogging thresholds shown as dashed lines. The steeper solid power / rotational speed curve represents the operating behavior of centrifugal pump 1 during clogging.

[0067] The described embodiments are merely examples that can be modified and / or supplemented in various ways within the scope of the claims. Each feature described for a particular embodiment can be used independently or in combination with other features in any other embodiment. Each feature described for an embodiment of a particular category can also be used accordingly in an embodiment of a different category. Reference symbol list

[0068] Wastewater pump 1 Pump housing 2 Suction opening 3 Wave 4 Motor 5 balance bike 6 steering 7 sensor 8 Acquisition of a temporal acceleration, pressure and / or motor current signal from the centrifugal pump 100 Fourier transforming the signal to obtain a frequency spectrum and / or a spectrogram 200 Analyzing the frequency spectrum and / or the spectrogram using a machine learning-trained image analysis algorithm to identify at least one spectral pattern for detecting the operating state. 300 Analyzing at least one spectral pattern with regard to a blockage as an operating condition of the centrifugal pump 400 Upon detection of a blockage, the spectral pattern is analyzed to classify the blockage of the centrifugal pump. 500 Clearing the blockage by operating the engine, depending on the classification of the blockage. 600

Claims

1. Method for detecting an operating state in a centrifugal pump (1) having a pump housing (2) with a shaft (4) arranged in the pump housing (2), an impeller (6) seated on the shaft (4) and a motor (5) driving the shaft (4), comprising the steps of: while pumping liquid with a possible solids content with the centrifugal pump (1), detecting a temporal acceleration, a pressure and / or a motor current signal of the centrifugal pump (1) (100), Fourier transforming the acceleration, the pressure and / or the motor current signal to obtain a frequency spectrum and / or a spectrogram (200), and analyzing the frequency spectrum and / or the spectrogram with regard to at least one spectral pattern to detect the operating state (300).

2. Method according to the preceding claim, wherein the analysis of the frequency spectrum and / or the spectrogram is carried out by means of an image analysis algorithm trained by machine learning.

3. Method according to one of the preceding claims, wherein the spectral pattern for detecting the operating state is characterized by exceeding an acceleration, pressure and / or motor current threshold value of an integer and / or half order of the acceleration, pressure and / or motor current signal.

4. Method according to one of the preceding claims, wherein the spectral pattern for detecting the operating state is characterized by exceeding a predefined time period, in particular a moving average value of the acceleration, pressure and / or motor current threshold value.

5. Method according to one of the preceding claims, comprising the step of analyzing the frequency spectrum and / or the spectrogram in a range between the 3rd and 9th order of the acceleration, pressure and / or motor current signal.

6. Method according to one of the preceding claims and according to claim 2, wherein the machine learning is carried out by means of a neural network, in particular a convolutional neural network, CNN.

7. Method according to one of the preceding claims and according to claim 2, comprising the step of analyzing the frequency spectrum and / or the spectrogram by means of the image analysis algorithm trained by machine learning with regard to several and / or different spectral patterns for detecting the operating state.

8. Method according to one of the preceding claims, comprising the steps of: analyzing the spectral pattern with regard to a blockage as an operating state of the centrifugal pump (1) (400), upon detection of the blockage, releasing the blockage by operating the motor (5) in at least one second direction of rotation opposite to a first direction of rotation for conveying the liquid.

9. Method according to the preceding claim, comprising the steps of: upon detection of the blockage, analyzing the spectral pattern with regard to a classification of the blockage of the centrifugal pump (500), and releasing the blockage by operating the motor (5) depending on the classification of the blockage (600).

10. Method according to one of the preceding claims, comprising the step: detecting the acceleration, pressure and / or motor current signal by means of an acceleration and / or vibration sensor (8), in particular by means of a 3-axis acceleration and / or vibration sensor, by means of a pressure sensor (8) and / or by means of a current sensor (8).

11. Method according to one of the preceding claims, comprising the step of: obtaining the frequency spectrum by means of fast Fourier transformation and / or the spectrogram by means of short-time Fourier transformation.

12. Method according to one of the preceding claims, comprising the step of: training the image analysis algorithm by means of a plurality of individual images with at least partially different operating states to obtain a respective vector with probabilities of a regular operating state and an abnormal operating state, in particular a non-blockage and a blockage.

13. Method according to the preceding claim, comprising the step of: training the image analysis algorithm when pumping clear water as a liquid and liquid with a defined solids content.

14. Method according to one of the preceding claims, comprising the steps: during the conveyance of liquid with the possible solids content, detecting a gradient and / or a curvature of a time profile of the acceleration, pressure and / or motor current signal of the centrifugal pump (1), and if the gradient and / or the curvature of the time profile of the acceleration, pressure and / or motor current signal exceeds a clogging threshold, Fourier transforming the signal.

15. Centrifugal pump (1) comprising a pump housing (2) with a shaft (4) arranged in the pump housing (2), an impeller (6) seated on the shaft (4), a motor (5) driving the shaft (4) and a controller (7), wherein the controller (7) is configured to detect a temporal acceleration, a pressure and / or a motor current signal of the centrifugal pump (1) while liquid with a possible solids content is being pumped by the centrifugal pump (1), to Fourier transform the acceleration, the pressure and / or the motor current signal to obtain a frequency spectrum and / or a spectrogram, and to analyze the frequency spectrum and / or the spectrogram with regard to at least one spectral pattern to detect an operating state of the centrifugal pump (1).

Citation Information

Patent Citations

  • Intelligent detection system for a pumping station.

    CH718585B1

  • Rotating machinery fault diagnosis method and system based on image fusion and integrated network

    CN115238785A

  • A method for monitoring operating condition of a rotating equipment

    EP3933355A1

  • Predictive maintenance using vibration analysis of vane pumps

    US20220318616A1

  • Detection of pump cavitation / blockage and seal failure via current signature analysis

    US7389278B2

Cited By

  • Low-temperature liquid cargo pump control and intelligent diagnosis method and system

    CN120273917A