A centrifugal pump fault diagnosis edge algorithm suitable for low-cost internet of things
By deploying a combined algorithm of db4 wavelet decomposition and fast Fourier transform at the centrifugal pump station in the oilfield, non-steady-state signals are processed in real time, solving the problem of limited cloud computing resources and realizing efficient diagnosis and resource optimization of centrifugal pump faults.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- LIAOHE GASOLINEEUM EXPLORATION BUREAU CO LTD
- Filing Date
- 2024-11-25
- Publication Date
- 2026-05-26
AI Technical Summary
Existing oilfield centrifugal pump fault diagnosis systems have high demands for non-steady-state signal processing when there are a large number of devices, resulting in excessive pressure on cloud computing resources and network bandwidth, affecting the accuracy and reliability of diagnosis, and the resources of edge devices are not fully utilized.
A combined algorithm of db4 wavelet decomposition and fast Fourier transform is deployed in the centrifugal pump station control system to process unsteady signals under equipment fault conditions in real time. Vibration signals are collected by accelerometers, and wavelet decomposition and Fourier transform are performed to extract low-frequency spectra and determine fault alarms.
It enables real-time detection of centrifugal pump faults, improves diagnostic accuracy, shortens fault detection and handling time, optimizes resource utilization, and reduces the risk of production interruption.
Smart Images
Figure CN122087291A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of instrumentation automation technology, specifically relating to an edge algorithm for centrifugal pump fault diagnosis suitable for low-cost Internet of Things. Background Technology
[0002] Centrifugal pumps are widely used fluid transport equipment in oilfield production. However, due to prolonged continuous operation, centrifugal pumps frequently experience various malfunctions, such as rotor misalignment and unbalanced vibration. These malfunctions affect the performance and reliability of the centrifugal pumps, and can even lead to production interruptions. Therefore, timely and accurate diagnosis of centrifugal pump malfunctions is crucial for ensuring the normal operation of oilfield production.
[0003] With the advancement of low-cost IoT projects and the promotion of unmanned station upgrades in the oilfield, Liaohe Oilfield has constructed numerous intelligent online monitoring systems for pumps and other moving equipment in various oil production plants. By deploying online monitoring devices and comprehensively utilizing data acquisition and industrial IoT technologies, the system achieves intelligent online monitoring of the operating status of each piece of equipment and equipment early warning identification. The main architecture involves deploying wireless temperature and vibration integrated sensors at the pump end in the field. The collected data is transmitted to a private cloud platform server via an intelligent gateway Ethernet transmission method. Intelligent algorithms deployed in the cloud are then used for computation to achieve fault diagnosis and location. While the existing system architecture has achieved online monitoring of pump fault conditions and effectively reduced the number of manual on-site inspections, it still has shortcomings. Under equipment fault conditions, such as unbalanced vibration in a centrifugal pump, a large number of non-steady-state signals are generated, accompanied by a large amount of noise. These signals have time-varying characteristics in the frequency domain, covering a wide frequency range with complex frequency components. When there are many devices, a large number of broadband signals need to be transmitted to the cloud for real-time processing and analysis, placing high demands on cloud computing resources and network bandwidth, affecting the accuracy and reliability of cloud data processing, and potentially causing the system to fail to respond to equipment faults in a timely manner. At the same time, there are also cases where the computing power of edge devices is not fully utilized, resulting in wasted resources. Therefore, processing and analyzing data at the station level, and using edge algorithm libraries to process non-steady-state signals under equipment failure conditions in real time, is of great significance for improving the efficiency and reliability of pump fault diagnosis.
[0004] The prior art ZL202110776005.3 discloses a method for diagnosing faults in centrifugal pumps of secondary water supply equipment. Although it effectively combines multiple signals to accurately judge the fault situation and improves the accuracy of fault judgment, it does not solve the problem of processing non-steady-state signals under equipment fault conditions. Summary of the Invention
[0005] To address the aforementioned problems, this invention proposes an edge algorithm for centrifugal pump fault diagnosis suitable for low-cost IoT, comprising the following steps:
[0006] S1. Set the sampling frequency and time window;
[0007] S2. Obtain the original data X n (t);
[0008] S3. Perform db4 wavelet decomposition on the original data;
[0009] S4. Perform Fast Fourier Transform (FFT) on the cA4 component.
[0010] Furthermore, in step S1, the radial vibration voltage signal of the accelerometer installed on the centrifugal pump bearing housing is collected, and the system sampling frequency and time window are set.
[0011] Further, in step S2, the discrete time-domain sequence X of the vibration signal is obtained. n (t) performs time-frequency domain analysis on the signal sequence to analyze the fault components.
[0012] Furthermore, in step S3, the time-domain signal is first subjected to Discrete Wavelet Transform (DWT), Mallat-tree decomposition, and single-layer one-dimensional Discrete Wavelet Transform.
[0013] Furthermore, the signal is decomposed layer by layer, and the approximate components of each layer are decomposed to obtain the approximate components and detail components of the next layer. The db4 wavelet basis function is selected. Finally, after four layers of decomposition of the signal, the wavelet decomposition coefficients C and data length L are obtained.
[0014] Furthermore, by combining the lengths of each data point in L with the wavelet decomposition coefficients C, the low-frequency approximate component cA4 is extracted.
[0015] Furthermore, in step S4, the frequency components of the extracted low-frequency approximation component cA4 are analyzed, and a Fast Fourier Transform (FFT) is used to obtain the low-frequency spectrum of the signal.
[0016] Furthermore, it also includes step S5: obtaining the rotor speed R (r / min) and extracting the signal amplitude value G at R / 60 (Hz).
[0017] Furthermore, it also includes step S6, determining if G is greater than or equal to a set threshold, and then outputting a fault alarm.
[0018] Furthermore, the process also includes determining if G is less than a set threshold, then moving the time window to retrieve the signal and repeating the above steps.
[0019] The beneficial effects of this invention are as follows:
[0020] This invention presents a real-time detection algorithm for unbalanced vibration faults in centrifugal pumps, which has many advantages and can achieve the following significant effects after implementation:
[0021] 1. This algorithm is based on mature signal analysis tools, has a rich program library, is easy to develop and maintain, and is suitable for low-cost IoT environments.
[0022] 2. By combining signal analysis algorithms, especially by using specific wavelet bases and multi-level decomposition, as well as fast Fourier transform for low-frequency spectrum analysis, the fault characteristics of rotor measuring points can be accurately extracted, greatly improving the accuracy of fault diagnosis.
[0023] 3. This algorithm can be deployed as an important component of the edge algorithm library within the control system of the station where the centrifugal pump is located, enabling real-time detection of unbalanced vibration faults in the centrifugal pump. It can quickly respond to changes in equipment status, greatly shorten the time for fault detection and handling, and effectively reduce the risk of production interruption caused by faults.
[0024] 4. By making full use of the station-level computing resources, the adverse effects caused by network transmission congestion and cloud computing resource shortages can be largely avoided, realizing the rational allocation and optimized utilization of resources, and becoming a powerful supplement to the existing cloud computing architecture pump fault detection system. Attached Figure Description
[0025] Figure 1 Here is a flowchart of a single-layer one-dimensional discrete wavelet transform.
[0026] Figure 2 This is a schematic diagram of the four-level decomposition of the original signal Xn(t) by the db4 wavelet;
[0027] Figure 3 Here is the algorithm flowchart;
[0028] Figure 4 This is a schematic diagram of the accelerometer's measuring points.
[0029] Figure 5 The original vibration signal data is shown in the graph (τ = 600ms).
[0030] Figure 6 The graph shows the original data of the vibration signal, the detail component D4, and the approximate component A4.
[0031] Figure 7 This is a frequency spectrum of the approximate component A4 and a fault alarm determination data graph. Detailed Implementation
[0032] Example 1
[0033] To make the technical means and objectives of this invention easier to understand, the invention is further described below with reference to specific embodiments. A centrifugal pump fault diagnosis edge algorithm suitable for low-cost IoT is mainly used to identify the unbalanced vibration fault state of centrifugal pumps, such as... Figure 3 As shown, the specific steps include the following:
[0034] S1. Set the sampling frequency and time window;
[0035] S2. Obtain the original data X n (t);
[0036] S3. Perform db4 wavelet decomposition on the original data;
[0037] S4. Perform Fast Fourier Transform (FFT) on the cA4 component.
[0038] In step S1, the radial vibration voltage signal of the accelerometer installed on the centrifugal pump bearing housing is collected, and the system sampling frequency and time window are set.
[0039] In step S2, the discrete time-domain sequence X of the vibration signal is obtained. n (t) performs time-frequency domain analysis on the signal sequence to analyze the fault components.
[0040] In step S3, the time-domain signal is first subjected to Discrete Wavelet Transform (DWT), Mallat-tree decomposition, and single-layer one-dimensional Discrete Wavelet Transform.
[0041] The signal is decomposed layer by layer, and the approximate components of each layer are decomposed to obtain the approximate components and detail components of the next layer. The db4 wavelet basis function is selected. Finally, after four layers of decomposition of the signal, the wavelet decomposition coefficients C and data length L are obtained.
[0042] Among them, the low-frequency approximate component cA4 is extracted from the wavelet decomposition coefficients C and combined with the length of each data in L.
[0043] In step S4, the frequency components of the extracted low-frequency approximation component cA4 are analyzed, and a Fast Fourier Transform (FFT) is used to obtain the low-frequency spectrum of the signal.
[0044] The process also includes step S5, obtaining the rotor speed R (r / min) and extracting the signal amplitude value G at R / 60 (Hz).
[0045] This also includes step S6, which determines if G is greater than or equal to a set threshold, and then outputs a fault alarm.
[0046] This includes a step to determine if G is less than a set threshold, then moving the time window to retrieve the signal and repeating the above steps.
[0047] Example 2
[0048] This invention patent research has developed a centrifugal pump fault diagnosis algorithm suitable for low-cost IoT. Deployed within the control system of the pump station, it aims to detect unbalanced vibration faults in centrifugal pumps in real time. It collects radial vibration voltage signals from the pump shaft end of the accelerometer, and extracts the low-frequency spectrum of the rotor measuring point through a combination of signal analysis algorithm tools. Based on the engineering practice characteristic that rotor imbalance will produce a vibration frequency equal to the rotational speed, and the amplitude is proportional to the amount of imbalance, it analyzes and determines whether there are fault components at the rotational speed frequency location, effectively detecting and diagnosing rotor imbalance faults, and thus drawing conclusions about the fault status.
[0049] This invention develops a diagnostic algorithm for unbalanced vibration faults in centrifugal pumps. It acquires radial vibration voltage signals from an accelerometer mounted on the centrifugal pump bearing housing, sets the system sampling frequency and time window, and obtains a discrete-time domain sequence of the vibration signal. Time-frequency domain analysis is then performed on the signal sequence to analyze the fault components. To ensure good frequency resolution and filter out interference frequencies from noise and singular signals, a Discrete Wavelet Transform (DWT) is first performed on the time-domain signal, followed by Mallat-tree decomposition. The single-layer one-dimensional discrete wavelet transform process is as follows: Figure 1 As shown.
[0050] The signal decomposition is performed layer by layer. Each layer's approximate components are decomposed to obtain the approximate and detail components for the next layer. The DB4 (Daubechies 4) wavelet basis function is selected. Finally, after four layers of signal decomposition, the wavelet decomposition coefficients C and data length L are obtained. The decomposition process is as follows: Figure 2 As shown.
[0051] From the wavelet decomposition coefficients C, and combined with the length of each data in L, the low-frequency approximate component cA4 is extracted, and its frequency components are analyzed. Here, Fast Fourier Transform (FFT) is used to obtain the low-frequency spectrum of the signal.
[0052] Obtain the pump rotor speed R (r / min), extract the signal amplitude value at R / 60 (Hz) from the spectrum, compare it with the set threshold, and output an alarm signal when the amplitude is greater than or equal to the set threshold; otherwise, do not alarm, continue to move the time window to obtain the signal, and repeat the above process.
[0053] Example 3
[0054] A horizontal single-stage cantilever centrifugal pump (flow rate 50 m³ / h) with an imbalance fault at a metering and transfer station of an oilfield. 3Taking a speed of 32m head and 2900r / min as an example, this algorithm will be explained:
[0055] 1. The sampling device is installed on the centrifugal pump bearing housing (e.g., Figure 4 The radial vibration voltage signal of the accelerometer at position (1) was sampled at a frequency of 3.2 kHz and a time window of 600 milliseconds was applied to obtain the discrete time domain sequence of the vibration signal.
[0056] The obtained raw data is as follows Figure 5 As shown.
[0057] The original vibration signal data underwent noise reduction and feature extraction. Discrete wavelet transform (DWT) was used to perform Mallat-tree decomposition on the signal, selecting the DB4 (Daubechies 4) wavelet basis function. After four levels of decomposition, the approximate component A4 was obtained (in actual programming, the decomposition coefficients are obtained by convolution and then downsampling the signal). The original signal and the detailed and approximate components of the four-level decomposition are as follows: Figure 6 As shown.
[0058] The frequency components of approximate component A4 are analyzed, and a Fast Fourier Transform (FFT) is performed to obtain the frequency spectrum. The pump speed is set to 1200 r / min, and the amplitude of the 20Hz frequency component is read and compared with the alarm threshold (0.03g). If the amplitude is greater than the alarm value, an alarm is triggered. The determination process is as follows: Figure 7 As shown.
[0059] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and concept of the present invention, should be covered within the scope of protection of the present invention.
Claims
1. An edge algorithm for centrifugal pump fault diagnosis suitable for low-cost Internet of Things, characterized in that, Includes the following steps: S1. Set the sampling frequency and time window; S2. Obtain the original data X n (t); S3. Perform db4 wavelet decomposition on the original data; S4. Perform Fast Fourier Transform (FFT) on the cA4 component.
2. The centrifugal pump fault diagnosis edge algorithm suitable for low-cost IoT as described in claim 1, characterized in that, In step S1, the radial vibration voltage signal of the accelerometer installed on the centrifugal pump bearing housing is collected, and the system sampling frequency and time window are set.
3. The centrifugal pump fault diagnosis edge algorithm suitable for low-cost IoT as described in claim 2, characterized in that, In step S2, the discrete time-domain sequence X of the vibration signal is obtained. n (t) performs time-frequency domain analysis on the signal sequence to analyze the fault components.
4. The centrifugal pump fault diagnosis edge algorithm suitable for low-cost IoT as described in claim 3, characterized in that, In step S3, the time-domain signal is first subjected to Discrete Wavelet Transform (DWT), Mallat-tree decomposition, and single-layer one-dimensional Discrete Wavelet Transform.
5. The centrifugal pump fault diagnosis edge algorithm suitable for low-cost IoT as described in claim 4, characterized in that, The signal is decomposed layer by layer, and the approximate components of each layer are decomposed to obtain the approximate components and detail components of the next layer. The db4 wavelet basis function is selected. Finally, after four layers of decomposition of the signal, the wavelet decomposition coefficients C and data length L are obtained.
6. The centrifugal pump fault diagnosis edge algorithm suitable for low-cost IoT as described in claim 5, characterized in that, From the wavelet decomposition coefficients C, and combined with the lengths of each data in L, the low-frequency approximate component cA4 is extracted.
7. The centrifugal pump fault diagnosis edge algorithm suitable for low-cost IoT as described in claim 6, characterized in that, In step S4, the frequency components of the extracted low-frequency approximation component cA4 are analyzed, and a Fast Fourier Transform (FFT) is used to obtain the low-frequency spectrum of the signal.
8. The centrifugal pump fault diagnosis edge algorithm for low-cost IoT as described in claim 7, characterized in that, It also includes step S5, obtaining the rotor speed R (r / min) and extracting the signal amplitude value G at R / 60 (Hz).
9. The centrifugal pump fault diagnosis edge algorithm for low-cost IoT as described in claim 8, characterized in that, It also includes step S6, which determines if G is greater than or equal to a set threshold, and then outputs a fault alarm.
10. The centrifugal pump fault diagnosis edge algorithm for low-cost IoT as described in claim 8, characterized in that, It also includes a step to determine if G is less than a set threshold, then move the time window to take the signal and repeat the above steps.