A drilling pump anomaly detection method based on multi-feature distance fusion and lateral contrast

Abstract

The application discloses the field of reciprocating mechanical state anomaly detection, and particularly relates to a drilling pump anomaly detection method based on multi-feature distance fusion and lateral comparison, which comprises the following steps: S1, collecting a reciprocating drilling pump fluid end vibration acceleration signal; S2, waveform abnormal vibration signal preprocessing; S3, calculating multi-dimensional time domain features and lateral comparison values; S4, calculating drilling pump fluid end abnormal state representation indexes; S5, calculating drilling pump fluid end abnormal state threshold values; and S6, deployment and application. The drilling pump anomaly detection model based on multi-feature distance fusion and lateral comparison calculates the degradation degree of different cylinder multi-dimensional time domain features of the vibration signal, quantifies the distance of the equipment offset health state center, enhances the sensitivity of the drilling pump fluid end degradation representation, and achieves the monitoring purpose of identifying the drilling pump fluid end abnormal state.

Description

Technical Field

[0001] This invention relates to the field of reciprocating mechanical condition anomaly detection technology, and in particular to a drilling pump anomaly detection method based on multi-feature distance fusion and lateral comparison. Background Technology

[0002] Reciprocating drilling pumps play a crucial role in oilfield development, serving as the core equipment of the drilling fluid circulation system and responsible for the continuous and stable delivery of drilling fluid to the bottom of the well. Their operational status directly affects the drilling fluid's effectiveness in carrying cuttings, cooling the drill bit, and stabilizing the wellbore. Any malfunction of the drilling pump can lead to unplanned downhole shutdowns, increased downhole complexity, higher energy consumption, and higher maintenance costs, even impacting the safety and efficiency of the entire drilling operation. Therefore, accurately and promptly identifying abnormal conditions in reciprocating drilling pumps is essential for ensuring the continuity and efficiency of oilfield production.

[0003] The core of anomaly identification for reciprocating drilling pumps lies in the learning and extraction of anomaly monitoring indicators. Oilfield drilling conditions are often complex and variable; pressure fluctuations, drilling fluid properties, and load changes all affect equipment operation. If the characteristic indicators are not extracted accurately, trend tracking will be biased, potentially leading to false or missed alarms regarding the drilling pump's operating status, adversely affecting the safety and stability of drilling operations. Therefore, this invention focuses on the problem of anomaly identification for reciprocating drilling pumps in the actual production environment of oilfield development, to meet the engineering application requirements for anomaly detection in equipment operation.

[0004] Existing anomaly identification methods often rely on a single characteristic value of equipment vibration, such as time-domain indicators like vibration intensity or kurtosis, to track its trend and identify and warn of anomalies based on set thresholds. The drawback is that the extracted single characteristic value only reflects one aspect of the drilling pump's operating condition and cannot comprehensively reveal the overall performance degradation of the equipment. When a drilling pump malfunctions, there is often a trend tracking bias, which can easily lead to false alarms or missed alarms, thereby increasing the operational and maintenance risks during oilfield development. Summary of the Invention

[0005] This invention provides a drilling pump anomaly detection method based on multi-feature distance fusion and lateral comparison. By calculating and extracting multi-dimensional drilling pump time-domain features, the method quantifies the degree of anomaly of the multi-dimensional feature vector deviating from the normal state, compensates for the insensitivity of single time-domain features to drilling pump anomalies, promptly detects the range of drilling pump anomaly deviations, and realizes early warning and maintenance of abnormal states of the hydraulic end of the drilling pump.

[0006] According to one aspect of this disclosure, a drilling pump anomaly detection method based on multi-feature distance fusion and lateral comparison is provided, the method comprising:

[0007] S1, collects vibration acceleration signals at the hydraulic end of a reciprocating drilling pump;

[0008] S2, Preprocessing of abnormal waveform vibration signals: Remove non-operational and strong non-periodic raw vibration acceleration signals;

[0009] S3, calculate the multidimensional time-domain features and the horizontal comparison value. The multidimensional time-domain features are the time-domain features of the original vibration signals of different cylinders at the hydraulic end of the drilling pump at different timestamps. The horizontal comparison value reflects the degree of difference in the changes between each cylinder.

[0010] S4, calculate the characterization index of abnormal state of the hydraulic end of the drilling pump, including: the m-dimensional characteristic ratio sequence of N cylinders obtained by step S3. and m-dimensional unit vector Calculate the Eulerian distance between the two. sequence;

[0011] S5, calculate the threshold for abnormal state of the hydraulic end of the drilling pump;

[0012] S6, Deployment and Application: Apply the abnormal state threshold of the equipment hydraulic end calculated in step S5 to the newly acquired real-time vibration data to perform abnormal state detection of the hydraulic end of the reciprocating drilling pump.

[0013] In one possible implementation, the hydraulic end measuring point includes multiple piston cylinders, and the vibration signal of each measuring point contains multi-channel vibration acceleration signals. The data model of the single-channel acceleration vibration signal of the measuring point is represented as follows: Where N is the number of cylinders, x i (t) represents the vibration acceleration signal of the i-th cylinder.

[0014] In one possible implementation, S2 includes: using a normalized autocorrelation function to measure the strong aperiodicity of the vibration signal, where the vibration signal removed given a running threshold is... Its normalized autocorrelation function is defined as:

[0015] ,in The number of sampling points. To delay, the denominator is energy normalization, which makes If a channel has a selected normalized autocorrelation function interval that satisfies If the signal is less than the empirical threshold, it is considered a strong aperiodic signal.

[0016] The strong aperiodic signal in S2 is mainly manifested in the presence of a single, significant impact component in the vibration signal.

[0017] In one possible implementation, S3 includes: assuming the number of cylinders in the hydraulic end of the drilling pump is N, calculating the time-domain characteristics of the original vibration signals of different cylinders in the hydraulic end of the drilling pump at different time stamps. Corresponding feature sequence The feature sequences include: root mean square (RMS), kurtosis, skewness, and odds center features;

[0018] Calculate the sliding timestamp Each cylinder The mean of the root mean square in the feature sequence is calculated, and the lateral contrast value sequence compared to other cylinders is calculated. A sliding time window is used to reflect the statistical characteristic changes of each cylinder, and a horizontal comparison value is used to reflect the degree of difference in changes between each cylinder. The mathematical model is as follows:

[0019] ,

[0020] Where m is the feature dimension. This represents the temporal feature sequence of the i-th cylinder on window k. Let be the mean of the root mean square of the j-th time-domain feature of the p-th cylinder on window k. This represents the average characteristic value of the remaining cylinders (excluding their own cylinder) within the sliding time window. Let be the m-dimensional feature ratio sequence of the i-th cylinder.

[0021] In one possible implementation, S4 includes: the Euclidean distance uses the L2 norm, i.e., the Euclidean distance, with the specific formula as follows:

[0022] ,

[0023] R is the set of real numbers.

[0024] In one possible implementation, S5 includes: calculating cylinder M. The 95th percentile of the sequence, multiplied by the mean and then by the trigger adjustment coefficient β, is used as the final detection threshold for hydraulic end anomalies in the early warning device. The threshold formula is as follows:

[0025] ,

[0026] Where q=95%, the Percentile function returns... 95% quantile of the sequence.

[0027] Compared with the prior art, the beneficial effects of the present invention are:

[0028] 1. This invention proposes a drilling pump anomaly detection method based on multi-feature distance fusion and lateral comparison. By comparing the lateral feature ratios of vibration signals from different cylinders, it avoids the problem of single-cylinder feature indicators failing due to random fluctuations and individual impact interference, thus improving the robustness of the detection method. Simultaneously, multi-feature fusion allows time-domain features such as RMS, kurtosis, skewness, and frequency center to complement each other, resulting in a more comprehensive characterization of the equipment's operational health status and effectively reducing the risk of misjudgment caused by single feature indicators. This method can reflect abnormal deviations in the hydraulic end based on multi-dimensional features.

[0029] 2. The drilling pump anomaly detection method proposed in this invention, which uses multi-feature distance fusion and lateral comparison, introduces Eulerian distance to quantify the deviation between the ratio of abnormal fluctuation features and the normal state. It then utilizes the feature distribution of statistical data obtained through a sliding time window and sets an anomaly trigger threshold, making the detection process practically effective. Furthermore, this method can not only be applied to drilling pumps but also extended to other types of reciprocating mechanical equipment, such as reciprocating compressors and reciprocating water injection pumps, achieving a unified anomaly monitoring framework across different equipment. Attached Figure Description

[0030] Figure 1 This is a flowchart of the present invention;

[0031] Figure 2 These are the vibration acceleration signals corresponding to the healthy and abnormal states of the dataset shown in the embodiment;

[0032] Figure 3 It is a sequence of hydraulic end vibration acceleration spectra of the dataset in healthy and abnormal states as shown in the embodiment;

[0033] Figure 4 The RMS trend of the vibration acceleration signal of cylinder 1 of the drilling pump shown in the example dataset;

[0034] Figure 5 This is the kurtosis trend of the vibration acceleration signal of cylinder 1 of the drilling pump in the dataset shown in the example;

[0035] Figure 6 This is the frequency center trend of the vibration acceleration signal of cylinder 1 of the drilling pump shown in the example dataset;

[0036] Figure 7 This is the skewness trend of the vibration acceleration signal of cylinder 1 of the drilling pump in the dataset shown in the example;

[0037] Figure 8 This is a horizontal comparison of the multi-feature distance quantization fusion of the three cylinders of the drilling pump shown in the embodiment, as well as its anomaly detection threshold. Detailed Implementation

[0038] Various exemplary embodiments, features, and aspects of this disclosure will now be described in detail with reference to the accompanying drawings. The same reference numerals in the drawings denote elements that have the same or similar functions. Although various aspects of the embodiments are shown in the drawings, they are not necessarily drawn to scale unless specifically indicated otherwise.

[0039] The term “exemplary” as used herein means “serving as an example, embodiment, or illustration.” Any embodiment illustrated herein as “exemplary” is not necessarily to be construed as superior to or better than other embodiments.

[0040] Furthermore, to better illustrate this disclosure, numerous specific details are set forth in the following detailed description. Those skilled in the art will understand that this disclosure can be practiced without certain specific details. In some instances, methods, means, components, and circuits well known to those skilled in the art have not been described in detail in order to highlight the main points of this disclosure.

[0041] Please refer to Figures 1-8 This invention provides a technical solution: a drilling pump anomaly detection method based on multi-feature distance fusion and lateral comparison, comprising the following steps:

[0042] S1. Acquiring Vibration Acceleration Signals from the Hydraulic End of a Reciprocating Drilling Pump: This involves acquiring vibration acceleration signals from the hydraulic end of a reciprocating drilling pump. The type of equipment used for signal acquisition is not limited to reciprocating drilling pumps; it is also applicable to other reciprocating mechanical equipment, such as reciprocating compressors and reciprocating water injection pumps. The hydraulic end measuring point includes multiple piston cylinders, and the vibration signal from each measuring point can contain multi-channel vibration acceleration signals. The data model for a single-channel acceleration vibration signal at a measuring point can be represented as follows: Where N is the number of cylinders, x i (t) represents the vibration acceleration signal of the i-th cylinder.

[0043] S2. Preprocessing of abnormal vibration signals: Removing non-operational and strongly aperiodic raw vibration acceleration signals. The key feature is the use of a normalized autocorrelation function to measure the strong aperiodicity of the vibration signal. Given an operating threshold, the removed vibration signal is... (Based on practical experience, an acceleration RMS threshold is set to remove non-operational vibration signals and retain vibration signals during equipment operation.) Its normalized autocorrelation function is defined as: ,in The number of sampling points. To delay, the denominator is energy normalization, which makes If a channel has a selected normalized autocorrelation function interval that satisfies If the signal is less than the empirical threshold, it is considered a strong aperiodic signal.

[0044] S3. Calculate multidimensional time-domain features and lateral comparison values: Assuming the number of cylinders in the hydraulic end of the drilling pump is N, calculate the time-domain original vibration signals of different cylinders in the hydraulic end of the drilling pump at different time stamps. Corresponding feature sequence Its features include RMS, kurtosis and skewness, but are not limited to these time-domain features.

[0045] Calculate the sliding timestamp Each cylinder The mean of the root mean square of the features and the lateral comparison value sequence relative to other cylinders are calculated. Its characteristic is that it uses a sliding time window to reflect the statistical characteristic changes of each cylinder, and uses a horizontal comparison value to reflect the degree of difference in changes between each cylinder. Its mathematical model is as follows:

[0046] ,

[0047] Where m is the feature dimension. This represents the temporal feature sequence of the i-th cylinder on window k. Let be the mean of the root mean square of the j-th time-domain feature of the p-th cylinder on window k. This represents the average characteristic value of the remaining cylinders (excluding their own cylinder) within the sliding time window. Let be the m-dimensional feature ratio sequence of the i-th cylinder.

[0048] S4. Calculate the characterization index of abnormal state of the hydraulic end of the drilling pump: use the m-dimensional characteristic ratio sequence of N cylinders obtained from step S3. Given an m-dimensional unit vector, calculate the Eulerian distance between them. The sequence is characterized by: the strong non-periodic vibration acceleration signal in S2 mainly manifests as a single, significant impact component in the vibration signal; the time-domain characteristics in S3... It includes characteristic indicators such as vibration acceleration RMS, acceleration kurtosis, and skewness, but is not limited to these time-domain indicators. The Euler distance in S4 uses the L2 norm, i.e., Euclidean distance, with the specific formula as follows:

[0049] .

[0050] S5. Calculate the threshold for abnormal hydraulic conditions at the drilling pump: Calculate cylinder N. The 50th percentile or higher of the sequence, multiplied by the mean and the trigger adjustment coefficient β, is used as the final detection threshold for hydraulic end anomalies in the early warning device. The threshold formula is as follows:

[0051] .

[0052] S6. Deployment and Application: Apply the equipment hydraulic end abnormal state threshold calculated in step S5 to the newly acquired real-time vibration data to perform hydraulic end abnormal state detection of reciprocating drilling pump.

[0053] To better illustrate the technical effects of the present invention, a specific example is used to verify the invention through experimentation. The data used in the embodiment is vibration operation data collected from the hydraulic end of a drilling pump in an oilfield. This dataset includes acceleration and vibration data under both healthy and abnormal conditions, such as... Figure 2 As shown, the corresponding spectrum data is as follows: Figure 3 As shown.

[0054] In this embodiment, an intelligent temperature and vibration sensor was used to collect hydraulic end vibration acceleration signals of the drilling pump over the past month, with a sampling frequency of 8000Hz. The RMS sequence, frequency center sequence, skewness sequence, and kurtosis sequence corresponding to the vibration acceleration signals were calculated respectively. Figures 4-7 As shown, a single time-domain feature cannot reflect the abnormal state of the hydraulic end of a drilling pump. The method of fusing multi-feature distance quantization and lateral comparison highlights the abnormal differences between cylinders in the hydraulic end. By using an abnormal state detection threshold, the deterioration state of the drilling pump's hydraulic end can be effectively predicted. Figure 8 As shown.

[0055] The various embodiments of this disclosure have been described above. These descriptions are exemplary and not exhaustive, nor are they limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is chosen to best explain the principles, practical application, or technical improvements to the embodiments in the market, or to enable others skilled in the art to understand the embodiments disclosed herein.

Claims

1. A method for drilling pump anomaly detection using multi-feature distance fusion lateral contrast, characterized in that, The method includes: S1, collects vibration acceleration signals at the hydraulic end of a reciprocating drilling pump; S2, Preprocessing of abnormal waveform vibration signals: Remove non-operational and strong non-periodic raw vibration acceleration signals; S3, calculate the multidimensional time-domain features and the horizontal comparison value. The multidimensional time-domain features are the time-domain features of the original vibration signals of different cylinders at the hydraulic end of the drilling pump at different timestamps. The horizontal comparison value reflects the degree of difference in the changes between each cylinder. S4, calculate the abnormal state characterization index of the hydraulic end of the drilling pump, including: the N cylinder m-dimensional characteristic ratio sequence calculated in S3 and the m-dimensional unit vector , calculate the Euler distance between the two sequence; S5, calculate the threshold for abnormal state of the hydraulic end of the drilling pump; S6, Deployment and Application: Apply the equipment hydraulic end abnormal state threshold calculated in step S5 to the newly acquired real-time vibration data to perform reciprocating drilling pump hydraulic end abnormal state detection. S3 comprises: assuming that the number of cylinders of the liquid end of the drilling pump is N, calculating the time-domain features of the original vibration signals of different cylinders of the liquid end of the drilling pump at different timestamps corresponding feature sequence , the feature sequence comprises: root mean square RMS, kurtosis, skewness and odds center features; Computing sliding timestamps Within each cylinder The mean of the root mean square of the feature sequence, and the lateral contrast value sequence compared with other cylinders Reflecting the statistical feature change of each cylinder with the sliding time window, and reflecting the difference degree of the change between each cylinder with the lateral contrast value, the mathematical model is: ， ， Where m is the feature dimension. This represents the temporal feature sequence of the i-th cylinder on window k. Let be the mean of the root mean square of the j-th time-domain feature of the p-th cylinder on window k. This represents the average characteristic value of the remaining cylinders (excluding their own cylinder) within the sliding time window. Let m be the sequence of feature ratios of the i-th cylinder; S4 includes: the Euclidean distance uses the L2 norm, i.e., the Euclidean distance, and the specific formula is: , R is the set of real numbers; S5 includes: calculating cylinder M. The 95th percentile of the sequence, multiplied by the mean and then by the trigger adjustment coefficient β, is used as the final detection threshold for hydraulic end anomalies in the early warning device. The threshold formula is as follows: Where q=95%, the Percentile function returns... 95% quantile of sequence; Where N is the number of cylinders.

2. The drilling pump anomaly detection method based on multi-feature distance fusion and lateral comparison according to claim 1, characterized in that, The hydraulic end measuring point includes multiple piston cylinders. The vibration signal of each measuring point contains multi-channel vibration acceleration signals. The data model of the single-channel acceleration vibration signal of the measuring point is represented as follows: Where N is the number of cylinders, x i (t) represents the vibration acceleration signal of the i-th cylinder.

3. The drilling pump anomaly detection method based on multi-feature distance fusion and lateral comparison according to claim 1, characterized in that, S2 includes: using a normalized autocorrelation function to measure the strong non-periodicity of the vibration signal, and removing the vibration signal after a given operating threshold. Its normalized autocorrelation function is defined as: ,in The number of sampling points. For the delay, the denominator is energy normalization, which makes If a channel has a selected normalized autocorrelation function interval that satisfies If the signal is less than the empirical threshold, it is considered a strong aperiodic signal. The strong aperiodic signal in S2 is mainly manifested in the presence of a single, significant impact component in the vibration signal.

Images