Intelligent monitoring method and system for operating state of mechanical pump

Abstract

The present application belongs to the technical field of mechanical pump operation monitoring, and provides a kind of mechanical pump's operating state intelligent monitoring method and system, comprising the following steps: obtaining the vibration signal and pressure signal of mechanical pump in the running process, vibration signal and pressure signal are respectively constructed as first time sequence and second time sequence, the positive transfer entropy from second time sequence to first time sequence is calculated, and the positive transfer entropy is used to quantify the causal influence intensity of pressure signal on vibration signal under the condition of known second time sequence history state.The present application has the unique ability to capture intermittent faults, even if the amplitude of vibration and pressure is large or small, high or low, but as long as the direction of causal relationship changes, the transfer entropy can capture this pulse type information flow anomaly, and the cumulative deviation overrun determines further ensure the stable identification of intermittent fault, solve the problem of intermittent fault false alarm.

Description

Technical Field

[0001] This invention belongs to the field of mechanical pump operation monitoring technology, specifically a method and system for intelligent monitoring of the operating status of mechanical pumps. Background Technology

[0002] As a core fluid transport device in industrial processes, mechanical pumps primarily function to convert mechanical energy into the pressure and kinetic energy of fluids through impeller rotation or volume changes, thereby achieving liquid transport, pressurization, or circulation. Common types of mechanical pumps include centrifugal pumps, reciprocating pumps, and gear pumps. They are the "heart" of the production process, and the stability and reliability of their operation directly affect the continuous production safety, operating efficiency, and energy consumption level of the entire process system. Because mechanical pumps are subjected to high-speed rotation, fluid impact, and complex loads for extended periods, key components (such as bearings, impellers, and seals) are highly susceptible to performance degradation or sudden failures. Among the many causes of mechanical pump failure, cavitation is one of the most common and destructive failure modes. Cavitation occurs when the pump inlet pressure is lower than the saturated vapor pressure of the liquid at the current temperature, causing the liquid to vaporize in the pump chamber and form a large number of bubbles. When these bubbles enter the high-pressure area with the liquid flow, they will collapse instantly and generate micro-jet and strong shock waves. The instantaneous pressure of this shock wave can reach hundreds or even thousands of atmospheres, which repeatedly act on the surface of the impeller, volute and other flow-through components, causing fatigue spalling of the metal material and forming honeycomb or sponge-like pits and holes. Cavitation not only reduces the pump's head and efficiency, but also causes severe vibration and noise, and may even directly lead to impeller damage or seal failure. Currently, traditional methods for monitoring and diagnosing cavitation in mechanical pumps mostly rely on periodic inspections or single vibration threshold alarms. However, in actual working conditions, the evolution of cavitation has obvious stage characteristics: in the very early stage of cavitation, due to the small number of bubbles and the energy of collapse concentrated in the high-frequency range, the pump's outlet pressure and overall vibration value often remain stable, making it difficult for traditional monitoring methods to detect abnormalities; as cavitation intensifies, the flow field is disturbed, and the outlet pressure begins to pulsate violently, but at this time the equipment may have already entered the middle and late stages of damage. If the monitoring strategy cannot be adaptively adjusted according to the stage characteristics of vibration and pressure signals, it is very easy to lead to the lack of early warning or misjudgment of faults. Once cavitation damage accumulates to a certain extent, it will not only cause equipment downtime for maintenance and increase operation and maintenance costs, but may also cause leakage of toxic and harmful media due to seal failure, resulting in serious production safety accidents and environmental pollution. Therefore, the present invention provides an intelligent monitoring method and system for the operating status of a mechanical pump. Summary of the Invention

[0003] In order to overcome the shortcomings of the prior art, at least one technical problem raised in the background art is solved.

[0004] The technical solution adopted by this invention to solve its technical problem is: an intelligent monitoring method for the operating status of a mechanical pump, comprising the following steps: The vibration and pressure signals of the mechanical pump during operation are acquired, and the vibration and pressure signals are constructed into a first time series and a second time series, respectively. Calculate the forward transfer entropy from the second time series to the first time series, wherein the forward transfer entropy is used to quantify the intensity of the causal influence of the pressure signal on the vibration signal under the condition that the historical state of the second time series is known; Simultaneously, the reverse propagation entropy from the first time series to the second time series is calculated, and the reverse propagation entropy is used to characterize the intensity of the causal influence of the vibration signal on the pressure signal. Based on the relationship between the forward and reverse transfer entropy, the dominant direction of energy transfer inside the mechanical pump at the current moment is determined, and based on the degree of deviation between the dominant direction and the preset health state benchmark direction, it is determined whether the mechanical pump has an early failure. If an early fault exists, the severity level of the fault is assessed based on the relationship between the forward and reverse propagation entropy, combined with the time-frequency domain characteristics of the vibration and pressure signals.

[0005] An intelligent monitoring system for the operating status of a mechanical pump, the system comprising: Signal acquisition module: acquires vibration and pressure signals of the mechanical pump during operation, and constructs the vibration and pressure signals into a first time sequence and a second time sequence, respectively; Intensity Analysis Module: Calculates the forward transfer entropy from the second time series to the first time series. The forward transfer entropy is used to quantify the intensity of the causal influence of the pressure signal on the vibration signal under the condition that the historical state of the second time series is known. Simultaneously, the reverse propagation entropy from the first time series to the second time series is calculated, and the reverse propagation entropy is used to characterize the intensity of the causal influence of the vibration signal on the pressure signal. Fault determination module: Based on the relationship between the forward and reverse transfer entropy, determine the dominant direction of energy transfer inside the mechanical pump at the current moment, and determine whether the mechanical pump has an early fault based on the degree of deviation between the dominant direction and the preset health status benchmark direction. Fault warning module: If an early fault exists, the severity level of the fault is assessed based on the relationship between the forward and reverse propagation entropy, combined with the time-frequency domain characteristics of the vibration and pressure signals.

[0006] The beneficial effects of this invention are as follows: This invention, through comparative analysis of bidirectional transfer entropy, can effectively distinguish whether the fault source originates from the fluid side or the solid side. When the forward transfer entropy of pressure → vibration decreases significantly while the reverse transfer entropy of vibration → pressure does not change significantly, it suggests that the fault may originate from the fluid side (e.g., insufficient inlet pressure). When the reverse transfer entropy increases significantly while the forward transfer entropy remains relatively stable, it suggests that the fault may originate from the solid side (e.g., reverse impact caused by bearing wear). When both bidirectional transfer entropies are abnormal, it suggests that there may be a complex coupled fault. This causal decoupling capability provides a clear direction for subsequent fault location and maintenance decisions. This invention has the unique ability to capture intermittent faults. Even if the amplitude of vibration and pressure fluctuates, as long as the direction of the causal relationship changes (i.e., the instantaneous collapse of the bubble destroys the determinism of energy transfer), the transfer entropy can capture this pulse-like information flow anomaly. The cumulative deviation exceeding the limit judgment (duration exceeding 3 windows) further ensures the stable identification of intermittent faults and solves the problem of missed intermittent faults. This invention not only enables fault monitoring and early warning, but also reveals the inherent physical laws of cavitation evolution. Under healthy conditions, the transfer of energy from fluid (pressure) to solid (vibration) is unidirectional and deterministic. When cavitation occurs, the random disturbance introduced by bubble collapse disrupts this unidirectional determinism, resulting in a reverse component in the information flow. As cavitation intensifies, the reverse information flow gradually strengthens until the dominant direction reverses. This quantitative characterization of the fault evolution process from an information theory perspective provides guidance for the condition assessment of mechanical pumps. Attached Figure Description

[0007] The invention will now be further described with reference to the accompanying drawings.

[0008] Figure 1 This is a flowchart of the steps of an intelligent monitoring method for the operating status of a mechanical pump according to the present invention; Figure 2 This is a module architecture diagram of an intelligent monitoring system for the operating status of a mechanical pump according to the present invention. Detailed Implementation

[0009] To make the technical means, creative features, objectives and effects of this invention easier to understand, the invention will be further described below in conjunction with specific embodiments.

[0010] Example 1 One of the core inventive points of this invention is that, in response to the problem that traditional mechanical pump monitoring methods are difficult to capture early cavitation signals and have a high rate of missed detection for intermittent faults, an intelligent monitoring method based on information entropy flow causal analysis is proposed. The core of this method is that, by collecting vibration and pressure signals in real time, the forward transfer entropy from pressure to vibration and the reverse transfer entropy from vibration to pressure are calculated to quantify the direction and intensity of information flow between the two. Instead of relying on the magnitude of the signal amplitude or the temporal relationship of fluctuations, the method judges whether early faults such as cavitation have occurred based on whether the dominant direction of information flow deviates from the unidirectional determinism (pressure → vibration) of the healthy state. This allows the method to capture weak disturbances and intermittent anomalies where the signal amplitude has not yet changed but the causal structure has been destroyed, thus achieving early warning. Please see Figure 1 As shown in the figure, the intelligent monitoring method for the operating status of a mechanical pump according to an embodiment of the present invention includes the following steps: Step 1: Acquire the vibration and pressure signals of the mechanical pump during operation, and construct the vibration and pressure signals into a first time series and a second time series, respectively; In step one, an acceleration vibration sensor is installed at the bearing housing of the mechanical pump body, and a pressure transmitter is installed near the pump body in the outlet pipeline of the mechanical pump. Vibration and pressure signals are simultaneously acquired at a sampling frequency of no less than 20kHz to ensure that the acquisition duration continuously covers multiple operating cycles of the pump. The acquired raw signals (vibration signals and pressure signals) are preprocessed, including: outlier removal, which can avoid false fluctuations caused by occasional interference from the sensor or bit errors in the data acquisition card; and frequency domain filtering, which filters out frequency components in the vibration and pressure signals that are not related to the physical mechanism of cavitation, and retains only the effective frequency bands that can characterize cavitation features, thereby improving the signal-to-noise ratio of subsequent transfer entropy calculation and the accuracy of fault identification. Specifically, frequency domain filtering includes the following processes: Vibration signal: Since the impact energy generated by cavitation is concentrated in the high-frequency band (usually 5kHz-20kHz), a bandpass filter is used to retain the information in this frequency band. The Butterworth filter is selected as the filter type (because its passband is flat and the transition band attenuation is controllable). The order is set to 4-6. The lower limit of the passband cutoff frequency is set to 5kHz-8kHz, and the upper limit is set to 18kHz-22kHz (which can be adjusted according to the actual sampling rate to ensure that the upper limit does not exceed the Nyquist frequency). The stopband attenuation is ≥40dB. Pressure signal: The main energy of the pressure signal is concentrated in the low frequency (usually below 500Hz). A low-pass filter is used to remove high-frequency noise. A Butterworth low-pass filter with an order of 4-6 is selected, and the cutoff frequency is set to 500Hz-1kHz (adjusted according to the pump speed and fluid pulsation characteristics). The stopband attenuation is ≥40dB. Filtering implementation: Zero-phase digital filtering (such as the filtfilt function) is used to avoid phase shift and ensure that the filtered signal is aligned with the original signal on the time axis; The preprocessed signal is segmented using a sliding time window. The window length L is set to include at least 10 pump rotation cycles. The rotation cycle is calculated based on the rated speed n of the mechanical pump as: T = 60 / n seconds. Therefore, the window length L ≥ 10T, where T is the time taken for the mechanical pump to rotate once. For example, for a mechanical pump with a rotation speed of 3000 rpm, T=0.02s, L≥0.2s, L=0.3s can be taken, and the window sliding step size is set to 10%-50% of the window length to achieve continuous monitoring; The vibration data within each window is constructed into a first time series, and the pressure data is constructed into a second time series; Zero-mean normalization is performed on the first and second time series within each window to eliminate the influence of amplitude dimensions, so that the transfer entropy calculation only reflects the information flow relationship between signals.

[0011] Step 2: Calculate the forward transfer entropy from the second time series to the first time series. The forward transfer entropy is used to quantify the degree to which the uncertainty of the future state of the first time series can be reduced under the condition that the historical state of the second time series is known, thereby characterizing the intensity of the causal influence of the pressure signal on the vibration signal. Simultaneously, the reverse propagation entropy from the first time series to the second time series is calculated to characterize the intensity of the causal influence of the vibration signal on the pressure signal. In step two, the transfer entropy, based on conditional entropy in information theory, is used to quantify the strength of the causal influence of one stochastic process on another. For the constructed first time series... Second time series Y= Forward transfer entropy from Y to X Defined as: ; in, This represents the value of the vibration sequence at time t+1 (future state). This represents the k-dimensional historical state vector (embedding dimension k) of the first time series at time t. This represents the pressure sequence at time t. 3D history state vector (embedding dimension) ); Denotes the joint probability density. Represents the conditional probability density; Similarly, the reverse propagation entropy from X to Y Defined as: ; in, This represents the value of the vibration sequence at time t+1 (future state). This represents the k-dimensional historical state vector (embedding dimension k) of the first time series at time t. This represents the pressure sequence at time t. 3D history state vector (embedding dimension) ); Denotes the joint probability density. Represents the conditional probability density; According to Takens' embedding theorem, for deterministic systems, the embedding dimension should satisfy k. ≥2d+1, where d is the supremum of the system's degrees of freedom. As a typical rotating machine, the dynamic behavior of a mechanical pump is mainly determined by its rotational frequency and its harmonics, resulting in relatively low degrees of freedom. In practical applications, the pseudo-nearest neighbor method is used to calculate historical data to determine k and... The value range of k is 3-5. In this embodiment, considering computational efficiency and information integrity, k is set to 5. =3; The prediction step size in the transfer entropy (defined as 1 in the above definition) represents the time interval for forward prediction. In this embodiment, the prediction step size is set to 1 to maximize the time resolution and capture transient causal relationships. The joint and conditional probabilities are calculated using the kernel density estimation method. A Gaussian kernel is chosen as the kernel function, and the bandwidth parameter is determined according to the Silverman rule. ; in, The standard deviation is represented by N, which represents the number of sample points. No prior assumptions about the data distribution are required, making it suitable for nonlinear and non-Gaussian mechanical signals. Within each sliding window, perform the following operations on the normalized first time series X and second time series Y: Based on the selected embedding dimension k= =3, construct the historical state vector as well as Where t ranges from 3 to N-1; All triples As sample points, estimate the joint probability density. Simultaneously estimate the conditional probability density ; Substitute the probability densities of all sample points into the transfer entropy formula and sum them to obtain the forward transfer entropy value from Y to X within the window. The unit is nant (base of natural logarithm) or bit (base of 2 logarithm). In this embodiment, natural logarithm is used, and the result is expressed in nant. Similarly, by constructing the relationship between future stress values ​​and historical states, the reverse propagation entropy value from X to Y is calculated: Normalize the propagation entropy. ;in, The conditional entropy represents the future state of vibration under given vibration history conditions. As a normalization factor, the reverse propagation entropy is also normalized. The normalized propagation entropy takes values ​​between 0 and 1, where 0 indicates no causal influence and 1 indicates completely deterministic causal driving.

[0012] For example, suppose that within a certain sliding window, the first time series X and the second time series Y are calculated to obtain... , This indicates that known pressure history significantly reduces the uncertainty of future vibration (0.35), while known vibration history has a weaker effect on reducing the uncertainty of future pressure (0.08). Therefore, the dominant direction of information flow in the current window is pressure → vibration, which is consistent with the characteristics of a healthy state. If subsequent windows show... Gradually increase or even exceed This indicates a change in the causal structure, which may indicate a fault such as cavitation.

[0013] Step two, by calculating the forward transfer entropy from pressure to vibration and the reverse transfer entropy from vibration to pressure, achieves a quantitative characterization of the energy transfer direction inside the mechanical pump and decoupling of the causal relationship. The transfer entropy can reveal the intrinsic driving relationship between pressure and vibration from the information theory perspective: when the mechanical pump is running healthily, fluid energy drives the solid response, and the dominant direction of information flow is pressure → vibration. When early failures such as cavitation occur, the random disturbances introduced by bubble collapse will disrupt this unidirectional deterministic energy transfer, leading to an increase in reverse transfer entropy or the appearance of bidirectional information flow. It can extract fault characteristics at an early stage by capturing minute shifts in the direction of information flow, even when the amplitude of vibration and pressure signals has not changed significantly and traditional threshold alarm methods have failed, thus solving the problem of weak and difficult-to-identify very early fault signals. As an amplitude-independent indicator, transfer entropy has a natural ability to resist interference from changes in operating conditions (such as speed adjustment and flow fluctuation), avoiding false alarms caused by changes in operating conditions. In addition, comparative analysis of bidirectional transfer entropy can also distinguish whether the fault source comes from the fluid side (abnormal pressure) or the solid side (abnormal vibration), providing a basis for subsequent fault location.

[0014] Step 3: Based on the relationship between the forward and reverse transfer entropy, determine the dominant direction of energy transfer inside the mechanical pump at the current moment, and based on the degree of deviation between the dominant direction and the preset health state benchmark direction, determine whether the mechanical pump has an early failure. In step three, the first step is to establish a baseline orientation for health status. The process is as follows: During the stable operation period of the mechanical pump, collect a sufficiently long period of health status data (it is recommended to collect data continuously for no less than 24 hours, covering different operating points of the pump), and calculate the forward and reverse transfer entropy of each window according to the methods in steps one and two above. For all windows in a healthy state, calculate the following statistics: mean and standard deviation of forward propagation entropy, mean and standard deviation of backward propagation entropy, and define the difference between the mean of forward propagation entropy and the mean of backward propagation entropy as the health baseline direction value; Under healthy conditions, since the normal working mechanism of a mechanical pump is that the average entropy of the forward transmission of fluid energy to drive the solid response is greater than the average entropy of the reverse transmission, that is, the dominant direction of information flow is pressure → vibration, this relationship is defined as the baseline direction of healthy conditions. During the online monitoring phase, as the sliding window moves forward, the real-time transfer entropy value of each window is continuously calculated; For each real-time window, calculate the real-time propagation entropy difference at the current moment; The dominant direction of the information flow in the current sliding window is determined based on the magnitude and sign of the real-time transmitted entropy difference. If the real-time entropy difference is greater than the dominant direction determination threshold, then the dominant direction is pressure → vibration (fluid-driven solid), which is consistent with the characteristics of a healthy state. If the absolute value of the real-time transmitted entropy difference is less than or equal to the threshold for determining the dominant direction, then the dominant direction is a two-way information flow (pressure and vibration influence each other), indicating that there may be a weak disturbance. If the real-time transmission entropy difference is less than the negative of the dominant direction determination threshold, then the dominant direction is vibration → pressure, indicating a significant anomaly in the causal structure. Among them, the dominant direction determination threshold The setting is based on the standard deviation of the positive propagation entropy under healthy conditions. Backward transfer entropy standard deviation Sure: ; in, This is the sensitivity coefficient, with a value ranging from 1 to 3; in this embodiment, it is set to 2. The difference between the health baseline orientation value and the real-time transmission entropy is calculated, and then the ratio of the difference to the health baseline orientation value is calculated. The percentage of the ratio is used as the causal deviation index. The difference between the real-time forward transfer entropy and the mean forward transfer entropy is calculated, and then the ratio of the difference to the standard deviation of the forward transfer entropy is calculated to obtain the positive statistical deviation. The difference between the real-time back propagation entropy and the mean back propagation entropy is calculated, and then the ratio of the difference to the standard deviation of the back propagation entropy is calculated to obtain the back statistical deviation. Based on the above indicators, if any of the following criteria are met, an early fault is determined to exist: Judgment condition 1: The real-time transmission entropy difference is less than the negative of the dominant direction judgment threshold; Judgment condition two: The causal deviation index is greater than or equal to the deviation index threshold and the duration exceeds 3 sliding windows; Judgment condition three: The positive statistical deviation is less than the first statistical deviation threshold and the negative statistical deviation is greater than the second statistical deviation threshold; Among them, the first statistical deviation threshold and the second statistical deviation threshold are opposites of each other, and the first statistical deviation threshold is negative. The setting is based on the 3σ principle of statistics and the trade-off consideration of engineering applications. Under healthy conditions, the fluctuations of forward and reverse propagation entropy usually follow a normal distribution. If none of the above three conditions are met, then it is determined that there is no early failure. It needs to be explained that the physical meaning of the above judgment conditions is as follows: When the real-time entropy difference is less than the negative of the dominant direction determination threshold, it indicates that the dominant direction of information flow has reversed from pressure→vibration in a healthy state to vibration→pressure. This is a clear sign of a fundamental change in the causal structure and directly indicates the presence of an early fault. This condition corresponds to the cavitation developing to a certain stage, where the reverse impact generated by bubble collapse is sufficient to change the dominant direction of energy transfer. When the causal deviation index is greater than or equal to the deviation index threshold, it indicates that the positive dominance has been lost by more than half and the information flow direction has seriously deviated from the healthy baseline, taking into account the possible short-term disturbances in the industrial site; When the forward statistical deviation is less than the first statistical deviation threshold and the reverse statistical deviation is greater than the second statistical deviation threshold, it indicates that the driving effect of pressure on vibration is significantly weakened, while the reverse effect of vibration on pressure is significantly enhanced. Even if the real-time transmission entropy has not yet reached the reversal threshold, it has already shown typical bidirectional information flow characteristics, and is judged to be an early fault.

[0015] Step 3 establishes a baseline direction for health status, calculates multi-dimensional indicators such as the difference in transmission entropy, causal deviation index, and statistical deviation in real time, and comprehensively utilizes three mutually synergistic fault identification logics: dominant direction reversal judgment, cumulative deviation exceeding limit judgment, and bidirectional statistical significant deviation judgment. This enables the system to keenly capture subtle changes in the causal structure of information flow and accurately identify early faults such as cavitation, even when the amplitude of vibration and pressure signals has not changed significantly and traditional threshold alarm methods have completely failed. At the same time, it effectively eliminates false alarms caused by random interference and operating condition fluctuations, providing a reliable basis for subsequent fault level assessment and early warning.

[0016] Step 4: If an early fault exists, assess the severity level of the fault based on the relationship between the forward and reverse propagation entropy, combined with the time-frequency domain characteristics of the vibration and pressure signals; generate early warning information based on the severity level of the fault. In step four, an analysis is performed based on a geometric distance-based evaluation method. The core of this method is to construct a feature space from multi-dimensional indicators, and the health status forms a health region in the space. The severity of the fault is determined by the distance of the current point from the center of the health region. During the healthy operation of the mechanical pump, M sliding windows are collected, and the following three-dimensional feature vectors are extracted from each window; ; in, This represents the ratio of bidirectional transfer entropy. This indicates the proportion of high-frequency energy in the vibration. Indicates the pressure pulsation coefficient; Calculate the mean vector and covariance matrix of the health feature vectors to construct the feature distribution of health status; Calculate the real-time feature vector extracted by the real-time window, and calculate the Mahalanobis distance from the real-time feature vector to the healthy distribution. ; It is worth explaining that the reason for using Mahalanobis distance is that it has a clear physical meaning, representing the standard deviation of the current state from the healthy state, follows a chi-square distribution, and can be used to calculate the significance level. Based on Mahalanobis distance Classify the severity level of the fault; like At this point, the current state deviates significantly from the healthy state (exceeding the 95% confidence interval), but has not yet reached the highly significant level (within the 99% confidence interval). This corresponds to the early stage of faults such as cavitation. The equipment performance may not have declined significantly yet, but the causal structure has begun to be destroyed, and it is judged as a minor fault. like At this point, the current state deviates significantly from the healthy state (exceeding the 99% confidence interval), the fault has developed to the middle and late stages, the information flow direction may have reversed, the high-frequency energy of vibration and pressure pulsation have increased significantly, and if not handled in time, it may cause equipment damage, and is judged as a severe fault; in, It is represented as the critical values ​​of a chi-square distribution with 3 degrees of freedom at 95% and 99% confidence levels, which can be found from the chi-square distribution table. Based on the significance levels commonly used in statistics, 95% is used as the warning line and 99% as the action line, which takes into account both the sensitivity and reliability of the early warning. While classifying the fault types, the time-frequency domain characteristics of vibration and pressure signals are used to assist in confirming the fault type and eliminate interference from other faults. If the vibration signal shows a wideband energy increase in the 5kHz-20kHz frequency band and the pressure signal shows high-frequency pulsation, then the fault type is confirmed to be cavitation. If the vibration signal shows a significant peak at the bearing's characteristic frequency (such as the failure frequency of the outer or inner ring), and the reverse propagation entropy change is not significant, it may be a bearing failure, and further judgment needs to be made in conjunction with other information. If the vibration signal energy increases significantly at the frequency of rotation, while the pressure signal remains stable, it may indicate an imbalance fault. If the auxiliary confirmation results do not match the cavitation, the fault label is adjusted according to the identification results, and the corresponding early warning strategy is adopted. Based on the severity level of the fault and the confirmed fault type, generate corresponding early warning information: Minor Fault Warning: A yellow warning is generated, indicating that "early signs of cavitation have been detected, the information flow direction has deviated, and the Mahalanobis distance is XX (exceeding the 95% confidence interval). It is recommended to pay attention to the inlet pressure, check the inlet filter status, and check the impeller during the next planned shutdown." The warning information is pushed to the operation and maintenance monitoring platform to remind operators to pay attention but no immediate intervention is required. Severe Fault Warning: A red warning is generated, indicating "Severe cavitation risk, causal structure has been reversed, Mahalanobis distance is XX (exceeding the 99% confidence interval), high-frequency vibration energy and pressure pulsation have increased significantly, continued operation may cause impeller damage or seal failure, it is recommended to stop the machine immediately for inspection or start the standby pump". If the system is equipped with an automatic interlock protection function, it can simultaneously send a stop command to the control system of the mechanical pump to trigger emergency stop protection. The early warning information includes the fault type, severity level, Mahalanobis distance value, confidence level, and recommended measures, and is recorded in the historical database for subsequent analysis and model optimization.

[0017] This embodiment can effectively distinguish whether the fault source comes from the fluid side or the solid side through comparative analysis of bidirectional transfer entropy. When the forward transfer entropy of pressure → vibration decreases significantly while the reverse transfer entropy of vibration → pressure does not change significantly, it suggests that the fault may originate from the fluid side (such as insufficient inlet pressure). When the reverse transfer entropy increases significantly and the forward transfer entropy is relatively stable, it suggests that the fault may originate from the solid side (such as reverse impact caused by bearing wear). When both bidirectional transfer entropies are abnormal, it suggests that there may be a complex coupling fault. This causal decoupling capability provides a clear direction for subsequent fault location and maintenance decisions. This embodiment has a unique ability to capture intermittent faults. Early faults such as cavitation often have intermittent characteristics, with bubbles appearing and disappearing intermittently, causing the signal to be sometimes good and sometimes bad. Traditional monitoring methods gradually lose the trust of maintenance personnel in the cycle of "alarm-reset-re-alarm". The special feature of this embodiment is that even if the amplitude of vibration and pressure fluctuates, as long as the direction of the causal relationship changes (i.e. the bubble collapse momentarily destroys the determinism of energy transfer), the transfer entropy can capture this pulse-like information flow anomaly. The cumulative deviation exceeding the limit judgment in step three (duration exceeding 3 windows) further ensures the stable identification of intermittent faults and solves the problem of missed intermittent faults that has long plagued the field of industrial monitoring. This embodiment not only realizes fault monitoring and early warning, but also reveals the inherent physical laws of cavitation evolution. Under healthy conditions, the transfer of energy from fluid (pressure) to solid (vibration) is unidirectional and deterministic. When cavitation occurs, the random disturbance introduced by bubble collapse disrupts this unidirectional determinism, resulting in a reverse component in the information flow. As cavitation intensifies, the reverse information flow gradually strengthens until the dominant direction reverses. This quantitative characterization of the fault evolution process from the perspective of information theory provides guidance for the condition assessment of mechanical pumps.

[0018] Example 2 Based on the same inventive concept as the intelligent monitoring method for the operating status of a mechanical pump in the foregoing embodiments, such as Figure 2 As shown, this application provides an intelligent monitoring system for the operating status of a mechanical pump, wherein the system specifically includes: Signal acquisition module: acquires vibration and pressure signals of the mechanical pump during operation, and constructs the vibration and pressure signals into a first time sequence and a second time sequence, respectively; Intensity analysis module: Calculates the transfer entropy from the second time series to the first time series. The transfer entropy is used to quantify the degree to which the uncertainty of the future state of the first time series can be reduced under the condition that the historical state of the second time series is known, thereby characterizing the intensity of the causal influence of the pressure signal on the vibration signal. Simultaneously, the reverse propagation entropy from the first time series to the second time series is calculated to characterize the intensity of the causal influence of the vibration signal on the pressure signal. Fault determination module: Based on the relationship between the forward and reverse transfer entropy, determine the dominant direction of energy transfer inside the mechanical pump at the current moment, and determine whether the mechanical pump has an early fault based on the degree of deviation between the dominant direction and the preset health status benchmark direction. Fault warning module: If an early fault exists, the severity level of the fault is assessed based on the relationship between the forward and reverse propagation entropy, combined with the time-frequency domain characteristics of the vibration and pressure signals; and warning information is generated based on the severity level of the fault.

[0019] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for intelligent monitoring of the operating status of a mechanical pump, characterized in that: Includes the following steps: The vibration and pressure signals of the mechanical pump during operation are acquired, and the vibration and pressure signals are constructed into a first time series and a second time series, respectively. Calculate the forward transfer entropy from the second time series to the first time series, wherein the forward transfer entropy is used to quantify the intensity of the causal influence of the pressure signal on the vibration signal under the condition that the historical state of the second time series is known; Simultaneously, the reverse transfer entropy from the first time series to the second time series is calculated, and the reverse transfer entropy is used to characterize the intensity of the causal influence of the vibration signal on the pressure signal. Based on the relationship between the forward and reverse transfer entropy, the dominant direction of energy transfer inside the mechanical pump at the current moment is determined, and based on the degree of deviation between the dominant direction and the preset health state benchmark direction, it is determined whether the mechanical pump has an early failure. If an early fault exists, the severity level of the fault is assessed based on the relationship between the forward and reverse propagation entropy, combined with the time-frequency domain characteristics of the vibration and pressure signals.

2. The intelligent monitoring method for the operating status of a mechanical pump according to claim 1, characterized in that: Acquire vibration and pressure signals, including: Simultaneously acquire vibration and pressure signals, and preprocess the acquired raw signals; The preprocessed signal is segmented using a sliding time window; The vibration data within each window is constructed into a first time series, and the pressure data is constructed into a second time series; Zero-mean normalization is performed on the first and second time series within each window.

3. The intelligent monitoring method for the operating status of a mechanical pump according to claim 1, characterized in that: The calculation process for the forward propagation entropy is as follows: For the constructed first time series Second time series Y= Forward transfer entropy from Y to X Defined as: ； in, This represents the value of the vibration sequence at time t+1. This represents the k-dimensional historical state vector (embedding dimension k) of the first time series at time t. This represents the pressure sequence at time t. 3D history state vector (embedding dimension) ); Denotes the joint probability density. Represents the conditional probability density; According to Takens' embedding theorem, determine the embedding dimensions k and l, where k = l = 3; The joint probability and conditional probability are calculated using the kernel density estimation method. A Gaussian kernel is selected as the kernel function, and the bandwidth parameter is determined according to the Silverman rule. Within each sliding window, a historical state vector is constructed for the normalized first time series X and second time series Y; By taking all triples as sample points, estimate the joint probability density and the conditional probability density. Substitute the probability densities of all sample points into the transfer entropy formula, sum them to obtain the forward transfer entropy value from Y to X within the window, and then normalize it.

4. The intelligent monitoring method for the operating status of a mechanical pump according to claim 1, characterized in that: The calculation process for the reverse propagation entropy is as follows: For the constructed first time series Second time series Y= Reverse transfer entropy from X to Y Defined as: ； in, This represents the value of the vibration sequence at time t+1; According to Takens' embedding theorem, determine the embedding dimensions k and l, where k = l = 3; The joint probability and conditional probability are calculated using the kernel density estimation method. A Gaussian kernel is selected as the kernel function, and the bandwidth parameter is determined according to the Silverman rule. Within each sliding window, a historical state vector is constructed for the normalized first time series X and second time series Y; By taking all triples as sample points, estimate the joint probability density and the conditional probability density. Substitute the probability densities of all sample points into the back propagation entropy formula, sum them to obtain the back propagation entropy value from X to Y within the window, and then normalize it.

5. The intelligent monitoring method for the operating status of a mechanical pump according to claim 1, characterized in that: The process of determining the dominant direction of energy transfer inside the mechanical pump at the current moment is as follows: During the stable operation period of the mechanical pump, health status data is collected, and the forward and reverse transfer entropy of each window are calculated to obtain the mean and standard of the forward transfer entropy, and the mean and standard deviation of the reverse transfer entropy. The difference between the mean of the forward transfer entropy and the mean of the reverse transfer entropy is defined as the health baseline direction value. During the online monitoring phase, the real-time forward propagation entropy and real-time reverse propagation entropy of each real-time window are calculated, and the difference in real-time propagation entropy is also calculated. If the real-time entropy difference is greater than the threshold for determining the dominant direction, then the dominant direction is pressure → vibration. If the absolute value of the real-time transmission entropy difference is less than or equal to the dominant direction determination threshold, then the dominant direction is a bidirectional information flow. If the real-time transmission entropy difference is less than the negative of the dominant direction determination threshold, then the dominant direction is vibration → pressure.

6. The intelligent monitoring method for the operating status of a mechanical pump according to claim 1, characterized in that: The process for determining whether a mechanical pump has an early-stage fault is as follows: Obtain the causal deviation index, positive statistical deviation, and negative statistical deviation; Based on the above indicators, if any of the following criteria are met, an early fault is determined to exist: The real-time transmission entropy difference is less than the negative of the dominant direction determination threshold; The causal deviation index is greater than or equal to the deviation index threshold and the duration exceeds 3 sliding windows; The positive statistical deviation is less than the first statistical deviation threshold and the negative statistical deviation is greater than the second statistical deviation threshold.

7. The intelligent monitoring method for the operating status of a mechanical pump according to claim 6, characterized in that: The calculation process for the causal deviation index is as follows: The difference between the health baseline orientation value and the real-time transmission entropy is calculated, and then the ratio of the difference to the health baseline orientation value is calculated. The percentage of the ratio is used as the causal deviation index.

8. The intelligent monitoring method for the operating status of a mechanical pump according to claim 6, characterized in that: The calculation process for the positive statistical deviation and the negative statistical deviation is as follows: The difference between the real-time forward transfer entropy and the mean forward transfer entropy is calculated, and then the ratio of the difference to the standard deviation of the forward transfer entropy is calculated to obtain the positive statistical deviation. The difference between the real-time backpropagation entropy and the mean backpropagation entropy is calculated, and then the ratio of the difference to the standard deviation of the backpropagation entropy is calculated to obtain the back statistical deviation.

9. The intelligent monitoring method for the operating status of a mechanical pump according to claim 1, characterized in that: The process for assessing the severity level of a fault is as follows: During the healthy operation of the mechanical pump, M sliding windows are collected, and the following three-dimensional feature vectors are extracted from each window; ； in, This represents the ratio of bidirectional transfer entropy. This indicates the proportion of high-frequency energy in the vibration. Indicates the pressure pulsation coefficient; Calculate the mean vector and covariance matrix of the health feature vectors to construct the feature distribution of health status; Calculate the real-time feature vector extracted by the real-time window, and calculate the Mahalanobis distance from the real-time feature vector to the healthy distribution. ; like The fault was determined to be minor. like The fault was determined to be severe.

10. An intelligent monitoring system for the operating status of a mechanical pump, characterized in that, The system is used to perform the method according to any one of claims 1-9, the system comprising: Signal acquisition module: acquires vibration and pressure signals of the mechanical pump during operation, and constructs the vibration and pressure signals into a first time sequence and a second time sequence, respectively; Intensity Analysis Module: Calculates the forward transfer entropy from the second time series to the first time series. The forward transfer entropy is used to quantify the intensity of the causal influence of the pressure signal on the vibration signal under the condition that the historical state of the second time series is known. Simultaneously, the reverse transfer entropy from the first time series to the second time series is calculated, and the reverse transfer entropy is used to characterize the intensity of the causal influence of the vibration signal on the pressure signal. Fault determination module: Based on the relationship between the forward and reverse transfer entropy, determine the dominant direction of energy transfer inside the mechanical pump at the current moment, and determine whether the mechanical pump has an early fault based on the degree of deviation between the dominant direction and the preset health status benchmark direction. Fault warning module: If an early fault exists, the severity level of the fault is assessed based on the relationship between the forward and reverse propagation entropy, combined with the time-frequency domain characteristics of the vibration and pressure signals.

Images