A method and system for intelligent analysis of valve performance test data

Abstract

This invention belongs to the field of data processing technology, specifically relating to an intelligent analysis method and system for valve performance test data. The method includes: constructing a dynamic sensitivity factor that reflects system stability by analyzing the rate of change of flow time series and the fluctuation of current time series, and generating a fatigue factor by combining pressure range; further, identifying and quantifying abnormal pressure peaks caused by water hammer effect, correcting the fatigue factor, and obtaining a degradation acceleration factor that better reflects real physical wear; finally, inputting this composite dynamic feature and traditional static feature together into a GBDT model for training, thereby capturing early failure signs and improving the accuracy and reliability of valve remaining service life prediction.

Description

Technical Field

[0001] This invention relates to the field of data processing technology. More specifically, this invention relates to an intelligent analysis method and system for valve performance test data. Background Technology

[0002] Valves are core components of industrial fluid control systems, widely used in critical fields such as energy, chemical industry, and aerospace. Their stability and reliability directly affect the safety and efficiency of the entire system; unexpected failures can lead to production interruptions or even serious safety accidents. Therefore, monitoring valve health status and predicting remaining service life to achieve predictive maintenance is of paramount practical significance for ensuring industrial production safety and reducing operation and maintenance costs.

[0003] To achieve accurate lifespan prediction, existing technologies typically collect multi-dimensional time-series data such as flow rate, pressure, and current of valves throughout their complete opening and closing cycles, and utilize advanced machine learning models for data mining and analysis. For example, Chinese patent document CN117034151B discloses a valve lifespan prediction method based on big data analysis, which uses a weighted random forest algorithm to predict valve lifespan by combining big data analysis with valve characteristics.

[0004] However, these methods typically rely on static features extracted from raw data, such as total switching time, peak pressure during testing, and total leakage rate—macroscopic statistical indicators—to build models. This approach ignores the deep physical information inherent in the dynamic changes of time-series data, such as minute fluctuations in drive current and the rate of change in flow curves. This dynamic information is precisely the key indicator of early valve wear, increased internal frictional resistance, or control system instability. Therefore, machine learning models based solely on static features struggle to capture early valve degradation trends, resulting in insufficient accuracy in their predictions and failing to meet the practical needs for high-precision predictive maintenance of critical equipment. Summary of the Invention

[0005] To address the aforementioned technical problem of insufficient accuracy in valve performance prediction, this invention provides solutions in the following aspects.

[0006] In a first aspect, the present invention provides an intelligent analysis method for valve performance test data, comprising:

[0007] A multidimensional time series for valve performance testing is obtained, including flow rate time series, current time series, inlet pressure time series, and outlet pressure time series. Based on the fluctuations of the flow rate and current time series, the valve's dynamic sensitivity factor is obtained. Combining the range of all pressure time series and the dynamic sensitivity factor, the valve's fatigue factor is obtained. Based on the abnormal pressure peak caused by water hammer, the valve's fatigue factor is corrected to obtain its degradation acceleration factor. A life prediction model is established based on the GBDT algorithm, the valve's degradation acceleration factor, and traditional static characteristics. The degradation acceleration factor and traditional static characteristics of the valve to be predicted are input into the life prediction model, and the predicted life of the valve is output, thus completing the valve's life prediction.

[0008] This invention constructs a degradation acceleration factor, integrates dynamic fluctuation information of flow and current reflecting system stability, reflects the load level pressure range, and corrects for abnormal pressure peaks caused by water hammer. By inputting this dynamic feature, which contains deep physical information, along with traditional static features into the GBDT model, the model can capture early valve wear and instability signs that are undetectable by existing technologies, thereby improving the accuracy and reliability of valve life prediction.

[0009] Preferably, the method for obtaining the test dynamic sensitivity factor of the valve includes:

[0010] Obtain the start and end times of each multidimensional time series, and denote any multidimensional time series as the target series; obtain the dynamic sensitivity factor of the valve in the target series; denote the mean of the dynamic sensitivity factors of the valve in all multidimensional time series as the test dynamic sensitivity factor of the valve.

[0011] This invention obtains a comprehensive index that fully reflects the overall operational stability of the valve throughout the entire test cycle by averaging the dynamic sensitivity factors of all multi-dimensional time series. This avoids evaluation bias caused by the randomness of a single test or the one-sidedness of specific working conditions, and provides a more representative input for the subsequent construction of a more reliable fatigue factor.

[0012] Preferably, the acquisition of the dynamic sensitivity factor of the valve in the target sequence includes:

[0013] Let the time interval of the target sequence be denoted as... Obtain the target current filtering sequence;

[0014] ;

[0015] In the formula, This represents the dynamic sensitivity factor of the valve in the target sequence; A time series representation of the flow rate of the target sequence; This represents the first derivative of the flow time series with respect to time; This represents the value of the first derivative of the flow time series with respect to time at the r-th time interval. The current time series of the target sequence represents the current value at the r-th time in the time interval. This represents the value of the target current filtering sequence at the r-th time in the time interval; Represents the absolute value function; This represents the normalization function.

[0016] This invention multiplies and integrates the absolute value of the flow rate change, which represents the valve's movement speed, with the degree to which the current value deviates from its mean, which represents the instability of the valve system. This amplifies the degradation signal of instability under high-speed movement and can more sensitively capture early signs of failure compared to analyzing any data sequence alone.

[0017] Preferably, obtaining the target current filter sequence includes: performing mean filtering on the current time series of the target sequence to obtain the target current filter sequence.

[0018] Preferably, the fatigue factor of the valve satisfies the expression:

[0019] ;

[0020] In the formula, The fatigue factor of the valve; The test dynamic sensitivity factor of the valve; Indicates the number of multidimensional time series; Represents the stress set of the i-th multidimensional time series; Represents the maximum value function; Describes the minimum value function; This represents the normalization function.

[0021] This invention establishes a fatigue model that is more in line with physical reality by multiplying the dynamic sensitivity factor of the test, which reflects the stability of the system, with the average pressure range, which reflects the overall load level. This model shows that the more unstable the system is, the greater the fatigue damage it will produce under the same load. Therefore, it can more comprehensively and accurately measure the comprehensive fatigue accumulation of the valve during the test than by considering only pressure or stability performance.

[0022] Preferably, the method for obtaining the degradation acceleration factor of the valve includes:

[0023] Obtain stable pressure; obtain the maximum peak value of water hammer shock waves in arbitrary multidimensional time series;

[0024] The degradation acceleration factor of the valve satisfies the following expression:

[0025] ;

[0026] In the formula, Indicates the degradation acceleration factor of the valve; The fatigue factor of the valve; This represents the maximum peak value of the water hammer shock wave in the i-th multidimensional time series. Represents the natural exponential function; This represents the normalization function.

[0027] This invention calculates the maximum peak value of the water hammer shock wave and uses this to exponentially correct the fatigue factor downwards. This effectively eliminates the exaggerated influence of short-term, severe fluid shocks on the actual mechanical fatigue assessment, making the final degradation acceleration factor more realistically reflect the wear degradation caused by continuous operation and improving the effectiveness of the feature.

[0028] Preferably, obtaining stable pressure includes: performing low-pass filtering on the pressure time series of all multidimensional time series to obtain several filtered pressure time series, and obtaining the mean of all filtered pressure time series, which is denoted as stable pressure.

[0029] Preferably, obtaining the maximum peak value of the water hammer shock wave of any multidimensional time series includes: taking the maximum value of the difference between the pressure set of the i-th multidimensional time series and the stable pressure as the maximum peak value of the water hammer shock wave of the i-th multidimensional time series.

[0030] Preferably, the establishment of the lifetime prediction model includes:

[0031] Construct a training sample set. For each valve sample used for training, construct a feature vector. The feature vector contains traditional static features and a degradation acceleration factor. The traditional static features include at least the total switching time, maximum test pressure, and leakage rate. The label of the valve sample is the actual lifespan.

[0032] Using the training sample set, a regression model is constructed and trained according to the existing standard steps of the GBDT algorithm to obtain a lifespan prediction model.

[0033] Secondly, the present invention provides an intelligent analysis system for valve performance test data, including a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the aforementioned intelligent analysis method for valve performance test data is implemented.

[0034] By adopting the above technical solution, a computer program is generated from the intelligent analysis method of valve performance test data and stored in a memory so that it can be loaded and executed by a processor. This allows for the creation of a terminal device based on the memory and processor, making it convenient to use.

[0035] The beneficial effects of this invention are as follows: Addressing the problem of inaccurate valve life prediction due to reliance on static features in existing technologies, a composite feature reflecting the dynamic degradation process of valves is constructed, a degradation acceleration factor, which assesses the stability of the valve system through dynamic fluctuations in flow and current; fatigue degree is calculated by combining pressure load, and the analysis interference caused by water hammer effect is corrected; by supplementing this dynamic feature with profound physical meaning into the traditional feature set, early failure signs that traditional methods cannot capture are provided for machine learning models, thereby improving the accuracy and reliability of valve life prediction. Attached Figure Description

[0036] Figure 1 This is a flowchart illustrating an intelligent analysis method for valve performance test data according to the present invention;

[0037] Figure 2 This is a schematic diagram showing the comparison of prediction effects before and after the introduction of the degradation acceleration factor. Detailed Implementation

[0038] This invention discloses an intelligent analysis method for valve performance test data, referring to... Figure 1 This includes steps S1-S4:

[0039] S1: Obtain a multidimensional time series for valve performance testing.

[0040] It should be noted that in order to fully capture the dynamic response characteristics of the valve, it is necessary to simultaneously collect multimodal data that can reflect the valve's state, such as flow data reflecting the fluid dynamics state, current data reflecting the drive system state, and pressure data reflecting the load state.

[0041] Specifically, a flow sensor, a current sensor, and a pressure sensor are deployed at the valve inlet. The flow sensor measures the fluid flow rate through the valve, the current sensor measures the drive current of the motor that drives the valve, and the pressure sensor measures the pressure at the valve inlet and outlet. During a complete valve opening and closing operation, time-series data from all sensors are synchronously collected at a preset sampling frequency. This time-series data is then converted into digitized time series using an analog-to-digital converter, resulting in several multidimensional time series of the valve. For example, the sampling frequency can be set to 10 seconds. To fully reflect changes in valve performance, the number of collected multidimensional time series can be relatively large, such as 1000.

[0042] Thus, several multidimensional time series of the valve have been obtained, including flow time series, current time series, inlet pressure time series, and outlet pressure time series.

[0043] S2: Based on the fluctuations of the flow rate time series and the current time series, obtain the valve's test dynamic sensitivity factor; combine the range of all pressure time series of the valve with the valve's test dynamic sensitivity factor to obtain the valve's fatigue factor.

[0044] It should be noted that existing prediction methods based on GBDT (Gradient Boosting Decision Tree) essentially equate the initial performance state of a valve with its future degradation trend, ignoring the deep physical information inherent in the dynamic response process that foreshadows early wear and fatigue accumulation. Therefore, this invention constructs a valve degradation acceleration factor, so that the model input is no longer limited to describing the static characteristics of the initial state, but includes an analysis of future degradation. Long-term valve wear is mainly determined by the physical fatigue the valve experiences during operation, which is determined by static and dynamic loads and the stability of the moving system. Static and dynamic loads are manifested in the magnitude of pressure, while the stability of the moving system is reflected in fluctuations in flow rate and current. Therefore, this invention first constructs a dynamic sensitivity factor to evaluate the stability of the moving system, and then constructs a fatigue factor by combining pressure data.

[0045] It should be further noted that the drive current of valve systems with internal friction or poor control will exhibit more minute fluctuations. Therefore, this invention analyzes the fluctuations in the valve's flow and current time series to obtain the valve's dynamic sensitivity factor. Furthermore, the valve's sensitivity to abnormal forces varies at different stages. For example, current spikes occurring during high-speed start-up and shutdown phases often indicate more severe internal resistance or instability compared to spikes of the same amplitude occurring during low-speed start-up and shutdown phases. Therefore, the greater the rate of change in the valve's flow data over a period of time, the greater the likelihood that the valve is in a high-speed start-up and shutdown phase, and thus the larger the corresponding dynamic sensitivity factor.

[0046] Specifically, based on the fluctuations of the flow rate time series and the current time series, the test dynamic sensitivity factor of the valve is obtained, including:

[0047] Obtain the start and end times of each multidimensional time series, denote any multidimensional time series as the target series, and denote the time interval of the target series as... .

[0048] The target current time series is mean filtered to obtain the target current filtered sequence.

[0049] The dynamic sensitivity factor of the valve in the target sequence satisfies the expression:

[0050] ;

[0051] In the formula, This represents the dynamic sensitivity factor of the valve in the target sequence; A time series representation of the flow rate of the target sequence; This represents the first derivative of the flow time series with respect to time; This represents the value of the first derivative of the flow time series with respect to time at the r-th time interval. The current time series of the target sequence represents the current value at the r-th time in the time interval. This represents the value of the target current filtering sequence at the r-th time in the time interval; Represents the absolute value function; This represents the normalization function.

[0052] In the formula, This represents the rate of change of the flow rate over a time series of the target sequence. Represents the absolute value of the rate of change of the flow rate in the time series of the target sequence; The larger the value, the greater the reliability of the rate of change of the valve's flow rate at the r-th time interval. This value represents the degree of current fluctuation of the target sequence's current time series at the r-th time in the time interval. The larger this value is, the greater the degree of current fluctuation of the valve at the r-th time in the time interval. This represents the contribution of the valve to the valve dynamic sensitivity factor at the r-th time in the time interval. The larger this value is, the greater the confidence that the valve has a large flow rate change rate at the r-th time in the time interval, and the greater the current fluctuation. Therefore, the greater the contribution of the valve to the valve dynamic sensitivity factor at the r-th time in the time interval. Indicates to The cumulative integral of the contribution of the valve to the dynamic sensitivity factor at all times within the target sequence. The larger this value is, the greater the dynamic sensitivity factor of the valve in the target sequence.

[0053] The mean of the dynamic sensitivity factors of the valve across all multidimensional time series is denoted as the test dynamic sensitivity factor of the valve.

[0054] Thus, the dynamic sensitivity factor of the valve was obtained.

[0055] It should be noted that the dynamic sensitivity factor reflects the stability of the valve system and is closely related to the degree of physical damage to the valve and the actual mechanical load it bears. Furthermore, under higher loads, the same energy fluctuations not only cause fluctuations in current and flow data but also generate more severe stress concentrations. The physical damage caused by an unstable system operating under high loads is far greater than that under low loads. Therefore, this invention combines the dynamic sensitivity factor with pressure data to establish a fatigue factor.

[0056] Preferably, the fatigue factor of the valve is obtained by combining the range of all pressure time series of the valve and the valve's test dynamic sensitivity factor, including:

[0057] It should be noted that the greater the average working pressure difference that the valve withstands during the test, the greater the level of static and dynamic mechanical load it bears. This indicates that the cumulative fatigue effect of the valve in a single operation is enhanced, and the risk of long-term life loss is correspondingly increased, resulting in a higher fatigue factor.

[0058] The fatigue factor of a valve satisfies the following expression:

[0059] ;

[0060] In the formula, The fatigue factor of the valve; The test dynamic sensitivity factor of the valve; Indicates the number of multidimensional time series; Represents the stress set of the i-th multidimensional time series; Represents the maximum value function; Describes the minimum value function; This represents the normalization function.

[0061] In the formula, This represents the pressure range of the valve in the i-th multidimensional time series, which represents the static and dynamic mechanical load levels borne by the valve. The larger this value is, the heavier the load on the valve in the corresponding period, and the higher the contribution of the pressure data of the i-th multidimensional time series to the valve's fatigue factor. This represents the average contribution of all multidimensional time series pressure data to the valve's fatigue factor; based on this, the larger the valve's dynamic sensitivity factor, the larger the valve's fatigue factor.

[0062] Thus, the fatigue factor of the valve was obtained.

[0063] S3: Based on the abnormal pressure peak caused by the water hammer effect, the fatigue factor of the valve is corrected to obtain the valve's degradation acceleration factor.

[0064] It should be noted that the fatigue factor of the S2 valve is calculated based on the valve's flow rate, current, and pressure data. However, this calculation method ignores the pervasive and highly destructive water hammer effect in valve systems. When the valve is rapidly closed in a liquid-filled pipeline, the fluid upstream of the valve continues to move forward due to inertia, and its kinetic energy is rapidly converted into pressure energy, forming an instantaneous high-pressure shock wave in front of the valve. This shock wave propagates back and forth in the pipeline at extremely high speeds, causing severe, high-amplitude, and brief spikes in the pressure data at the end of the valve's operation. If the pressure data containing these spurious spikes is directly averaged to calculate the average working pressure difference, these high-amplitude instantaneous spikes will disproportionately inflate the average value, leading to spurious fluctuations in the pressure difference data. Consequently, the calculated fatigue factor cannot accurately reflect the valve's true fatigue condition. Therefore, this invention corrects for the errors caused by the water hammer effect.

[0065] It should be further explained that the larger the water hammer peak value, the more severe the disturbance to the fluid system when the valve is closed, and the greater the instantaneous impact load it bears. Therefore, the difference between the pressure affected by water hammer and the pressure under actual steady-state conditions can characterize the strength of the water hammer. The larger the water hammer peak value, the smaller the actual steady-state pressure, and the more the valve's fatigue factor should be corrected downwards, thus obtaining the valve's degradation acceleration factor.

[0066] Specifically, low-pass filtering is performed on all multidimensional time series pressure time series to obtain several filtered pressure time series. The mean of all filtered pressure time series is then obtained and denoted as the stable pressure.

[0067] Preferably, the degradation acceleration factor of the valve satisfies the expression:

[0068] ;

[0069] ;

[0070] In the formula, Indicates the degradation acceleration factor of the valve; The fatigue factor of the valve; This represents the maximum peak value of the water hammer shock wave in the i-th multidimensional time series. Represents the stress set of the i-th multidimensional time series; Indicates stable pressure; Represents the maximum value function; Represents the natural exponential function; This represents the normalization function.

[0071] In the formula, The maximum value of the difference between the pressure set of the i-th multidimensional time series and the steady pressure is represented by the maximum peak value of the water hammer shock wave. The larger this value is, the smaller the correction degree of the fatigue factor of the valve should be, and the larger the degradation acceleration factor of the valve should be. This represents the cumulative maximum peak value of the water hammer shock wave reflected in all multidimensional time series. The larger this value is, the lower the overall correction degree should be, and therefore the greater the degradation acceleration factor of the valve.

[0072] Thus, the degradation acceleration factor of the valve was obtained.

[0073] S4: Based on the GBDT algorithm and the valve degradation acceleration factor, a life prediction model is established to predict the valve's life.

[0074] It should be noted that this invention uses the valve degradation acceleration factor as an innovative feature to supplement the traditional feature set, and together they are used to train a more accurate life prediction model.

[0075] Specifically, a training sample set is constructed. For each valve sample used for training, a feature vector is constructed. The feature vector contains traditional static features and a degradation acceleration factor. The traditional static features include at least the total switching time, the maximum test pressure, and the leakage rate. The label of the valve sample is the actual lifespan.

[0076] Using the aforementioned training sample set, a regression model is constructed and trained according to the existing standard steps of the GBDT algorithm to obtain a lifetime prediction model. It should be noted that the GBDT model achieves high-precision prediction by iteratively constructing a series of weak decision trees, each new tree aiming to correct the residuals of all previous trees. By incorporating a degradation acceleration factor into the model input, GBDT can utilize this strong feature to find the critical path distinguishing between long and short lifetimes at an earlier stage with fewer splits when constructing the decision tree, thus capturing early failure signs that traditional methods cannot identify.

[0077] Input the feature vector of the valve to be predicted into the life prediction model, and output the predicted life of the valve. For example... Figure 2 A comparison chart showing the prediction effects before and after the introduction of the degradation acceleration factor.

[0078] This completes the prediction of the valve's lifespan.

[0079] This invention also discloses an intelligent analysis system for valve performance test data, including a processor and a memory. The memory stores computer program instructions, and when the computer program instructions are executed by the processor, an intelligent analysis method for valve performance test data according to this invention is implemented.

[0080] The system also includes other components well known to those skilled in the art, such as communication buses and communication interfaces, the settings and functions of which are known in the art and will not be described in detail here.

[0081] While this specification has shown and described numerous embodiments of the invention, it will be apparent to those skilled in the art that such embodiments are provided by way of example only. Many modifications, alterations, and alternatives will occur to those skilled in the art without departing from the spirit and essence of the invention. It should be understood that various alternatives to the embodiments of the invention described herein may be employed in the practice of this invention.

Claims

1. An intelligent analysis method for valve performance test data, characterized in that, include: A multidimensional time series for valve performance testing is obtained, comprising flow rate time series, current time series, inlet pressure time series, and outlet pressure time series. Based on the fluctuations of the flow rate time series and the current time series, the test dynamic sensitivity factor of the valve is obtained, including: obtaining the start and end times of each multidimensional time series, denoteing any multidimensional time series as the target series, and denoteing the time interval of the target series as... The target current time series is mean-filtered to obtain the target current filtered series; the dynamic sensitivity factor of the valve in the target series satisfies the expression: , This represents the dynamic sensitivity factor of the valve in the target sequence. A time series representation of the flow rate of the target sequence; This represents the first derivative of the flow time series with respect to time, and also represents the rate of change of the flow in the target sequence's flow time series; This represents the value of the first derivative of the flow time series with respect to time at the r-th time interval. This represents the current value at the r-th time in the time interval of the target sequence's current time series. This represents the value of the target current filtering sequence at the r-th time in the time interval. This represents the degree of current fluctuation in the target sequence's current time series at the r-th time point within the specified time interval. Represents the absolute value function. The normalization function is represented; the mean of the dynamic sensitivity factors of the valve across all multidimensional time series is denoted as the test dynamic sensitivity factor of the valve. By combining the range of all pressure time series of the valve and the test dynamic sensitivity factor of the valve, the fatigue factor of the valve is obtained. Based on the abnormal pressure peak caused by water hammer effect, the fatigue factor of the valve is corrected to obtain the valve degradation acceleration factor. Based on the GBDT algorithm, the degradation acceleration factor of the valve, and traditional static characteristics, a life prediction model is established. The degradation acceleration factor and traditional static characteristics of the valve to be predicted are input into the life prediction model, and the predicted life of the valve to be predicted is output, thus completing the life prediction of the valve.

2. The intelligent analysis method for valve performance test data according to claim 1, characterized in that, The fatigue factor of the valve satisfies the following expression: ； In the formula, The fatigue factor of the valve; The test dynamic sensitivity factor of the valve; Indicates the number of multidimensional time series; Represents the stress set of the i-th multidimensional time series; Represents the maximum value function; Describes the minimum value function; This represents the normalization function.

3. The intelligent analysis method for valve performance test data according to claim 1, characterized in that, The method for obtaining the degradation acceleration factor of the valve includes: Obtain stable pressure; obtain the maximum peak value of water hammer shock waves in arbitrary multidimensional time series; The degradation acceleration factor of the valve satisfies the following expression: ； In the formula, Indicates the degradation acceleration factor of the valve; The fatigue factor of the valve; This represents the maximum peak value of the water hammer shock wave in the i-th multidimensional time series. Represents the natural exponential function; This represents the normalization function.

4. The intelligent analysis method for valve performance test data according to claim 3, characterized in that, The process of obtaining stable pressure includes: performing low-pass filtering on the pressure time series of all multidimensional time series to obtain several filtered pressure time series, and obtaining the mean of all filtered pressure time series, which is denoted as stable pressure.

5. The intelligent analysis method for valve performance test data according to claim 3, characterized in that, The method of obtaining the maximum peak value of water hammer shock wave in any multidimensional time series includes: taking the maximum value of the difference between the pressure set of the i-th multidimensional time series and the stable pressure as the maximum peak value of water hammer shock wave in the i-th multidimensional time series.

6. The intelligent analysis method for valve performance test data according to claim 1, characterized in that, The establishment of the lifetime prediction model includes: Construct a training sample set. For each valve sample used for training, construct a feature vector. The feature vector contains traditional static features and a degradation acceleration factor. The traditional static features include at least the total switching time, maximum test pressure, and leakage rate. The label of the valve sample is the actual lifespan. Using the training sample set, a regression model is constructed and trained according to the existing standard steps of the GBDT algorithm to obtain a lifespan prediction model.

7. An intelligent analysis system for valve performance test data, characterized in that, include: A processor and a memory, wherein the memory stores computer program instructions that, when executed by the processor, implement an intelligent analysis method for valve performance test data according to any one of claims 1-6.

Images