Physical sensor driven pump valve health assessment method

Abstract

The invention relates to the technical field of mechanical structure health monitoring and diagnosis, and discloses a health assessment method for a pump valve driven by a mechanical sensor. A plurality of acceleration sensors are axially arranged on a pump valve shell, static sampling is firstly carried out to solve a static mean value, and zero offset correction is carried out on an operation sequence; setting a unified sampling period and time length, and synchronously acquiring synthetic vibration averaged according to moments; establishing discrete frequency coordinates, and calculating sine and cosine superposition amplitudes frequency by frequency to form a spectrum; performing neighborhood comparison and peak taking, calculating an energy ratio, and determining main resonance by using the energy with the maximum ratio and the smaller frequency; setting sliding windows, and solving dominant frequency window by window; drifting is calculated with the first window as the reference and normalized into a relative drifting ratio; constructing main frequency stability according to the drift difference and the drift distance of the adjacent windows, and calculating a mean value; and constructing health indexes according to the relative drift ratio average absolute value and the stability mean value, and performing graded output.

Description

Technical Field

[0001] This invention relates to the field of mechanical structure health monitoring and diagnosis technology, specifically a pump and valve health assessment method driven by a mechanical sensor. Background Technology

[0002] With the large-scale application of pumps and valves in key industrial sectors such as petroleum, chemical, and power, pump and valve health monitoring and fault early warning have become core aspects of ensuring production continuity and safety. Current technologies for assessing pump and valve health primarily rely on periodic manual inspections, experience-based judgment, or indirect analysis using traditional parameters (such as pressure, flow rate, and temperature). However, these methods generally suffer from limitations such as poor real-time performance, high subjectivity, and difficulty in quantifying structural damage, failing to meet the urgent needs of modern industry for online, accurate, and quantitative assessment of equipment operating status.

[0003] In recent years, with the development of mechanical sensor technology and signal analysis algorithms, some studies have begun to attempt to use accelerometers to collect vibration signals of pump and valve structures and identify operational anomalies through spectral analysis. However, existing signal analysis methods mostly employ conventional frequency domain algorithms such as standard Fourier transform or short-time Fourier transform, which have limited ability to capture the drift and dynamic changes of the main resonant frequency, making it difficult to reveal early damage and health degradation trends caused by fatigue, loosening, and cracks within the equipment. Especially under complex operating conditions, traditional methods are relatively crude in accurately tracking the main frequency peak and dynamically attributing multi-window drift, and are easily affected by external interference such as noise, load fluctuations, and installation processes, resulting in insufficient stability and sensitivity of monitoring results. In addition, most current mechanical health monitoring methods rely heavily on manually setting thresholds and rules or introducing prior models, which are insufficiently adaptable to new pump and valve structures and variable load conditions. Once there is a deviation between the actual operating environment and the model settings, it is easy to lead to misjudgment or missed judgment. Meanwhile, existing technologies often lack the detailed processing for time window subdivision, sliding analysis, and normalization evaluation of the dynamic characteristics of the frequency domain main peak, making it difficult to quantitatively track and evaluate the stability of the continuous evolution of the pump and valve structure's main frequency over time. This makes it difficult to identify subtle changes in the early stages of fault evolution in a timely manner, affecting the scientific and intelligent level of pump and valve life cycle health management.

[0004] Therefore, this case aims to propose a mechanical sensor-driven pump and valve health assessment method. By deploying multiple acceleration sensors along the axial direction on the pump and valve housing, multi-channel high-resolution acquisition of the overall structural vibration response is achieved. Combined with multi-level algorithms such as static calibration, spatiotemporal correction, frequency domain decomposition, and time-varying dominant frequency tracking, key health characteristics such as the pump and valve's main resonant frequency, dominant frequency drift, and vibration stability can be extracted in real time and accurately. Finally, the structural state classification results are output in the form of a health index. Summary of the Invention

[0005] This invention provides a pump and valve health assessment method driven by a mechanical sensor, which helps to solve the problems mentioned in the background art.

[0006] This invention provides the following technical solution: a pump valve health assessment method driven by a mechanical sensor, comprising: Multiple acceleration sensors are axially installed on the pump and valve housing. Static sampling is performed, and the static average value of each sensor is calculated. Based on the static average value, zero-bias correction is performed on the running sampling sequence to obtain the correction sequence. Set a uniform sampling period and total duration, synchronously collect the calibration sequences of each sensor based on a uniform clock, perform an arithmetic average at each sampling moment, and generate a synthetic vibration sequence; Establish discrete frequency coordinates, and calculate the superimposed amplitude of sine and cosine waves around each discrete frequency of the synthesized vibration sequence to form an amplitude spectrum; Perform neighborhood comparisons within the amplitude spectrum to obtain a set of peaks, calculate the energy and energy percentage of each peak, and determine the main resonance frequency based on the peak with the largest energy percentage and the smallest frequency value. Set the sliding time window and step size, extract the window signals in sequence and calculate the window amplitude spectrum to obtain the main resonance frequency of each window; The main resonance frequency of the first window is selected as the benchmark, and the drift of the main resonance frequency of each window is calculated to generate a relative drift ratio sequence. Calculate the ratio of the relative drift difference to the drift amount combination between two adjacent windows to obtain the main frequency stationarity of each window and calculate the mean stationarity. Calculate the average absolute value of the relative drift ratio, construct a health index by combining it with the mean of the stability, complete the classification based on the threshold, and output the main resonance frequency, drift amount, relative drift ratio, main frequency stability, mean stability, average absolute value of drift, and health index.

[0007] Optionally, the step of axially mounting multiple acceleration sensors on the pump valve housing, performing static sampling, calculating the static average value of each sensor, and performing zero-bias correction on the running sampling sequence based on the static average value to obtain a correction sequence specifically includes: Select multiple installation points at equal intervals along the axis of the pump and valve housing and number them. Install piezoelectric accelerometers and record the correspondence between the installation points and the sensor numbers. Under conditions where the equipment is stationary and pressureless, perform static sampling of all sensors for a uniform duration to obtain the static sampling sequence of each sensor; The arithmetic mean of the static acceleration for each sensor is calculated as the zero bias parameter; The original acceleration sequence was collected under the operating conditions of the equipment, and each time sample was calibrated according to the zero bias parameter to obtain the calibrated acceleration sequence of each sensor.

[0008] Optionally, the step of setting a unified sampling period and total duration, synchronously acquiring the calibration sequences of each sensor based on a unified clock, and performing an arithmetic mean at each sampling moment to generate a synthetic vibration sequence specifically includes: Set a uniform sampling period and total sampling duration, and calculate the number of samples to be collected; The discrete time series of calibrated acceleration is obtained for each sensor according to a unified clock. At each discrete time step, the calibration values ​​of all sensors are arithmetically averaged to generate a synthetic vibration sequence.

[0009] Optionally, the step of establishing discrete frequency coordinates and calculating the superimposed sine and cosine amplitudes of the synthesized vibration sequence around each discrete frequency to form an amplitude spectrum specifically includes: The sampling frequency is calculated based on the sampling period, and the set of discrete frequencies up to the Nyquist frequency is determined. For each frequency in the discrete frequency set, the superposition intensity of the cosine component and the superposition intensity of the sine component are calculated for the synthesized vibration sequence, and the amplitude of the corresponding frequency point is obtained based on the combination of the two. The output covers the amplitude spectrum of all discrete frequencies.

[0010] Optionally, the step of performing neighborhood comparison within the amplitude spectrum to obtain a peak set, calculating the energy and energy percentage of each peak, and determining the main resonance frequency according to the peak with the largest energy percentage and the smallest frequency value specifically includes: Perform left and right neighborhood amplitude comparisons at comparable in-place points of the amplitude spectrum to generate a set of candidate peaks; For each peak in the candidate peak set, calculate the peak energy by squared amplitude, and sum the energy across the entire frequency band; Peak identification is terminated when the energy across the entire frequency band is zero or the candidate peak set is empty. Calculate the energy percentage of each peak; The main resonance frequency is determined by the peak with the largest energy percentage. When the energy percentages are equal, the frequency corresponding to the peak with the smaller frequency value is selected.

[0011] Optionally, the step of setting the sliding time window and step size, extracting the window signals sequentially and calculating the window amplitude spectrum to obtain the main resonance frequency of each window specifically includes: Set the sliding time window length and step size, convert them to integer sample numbers, and calculate the number of windows; Calculate the start and end indices of each window according to its sequence number, and extract the window signal; Establish a discrete frequency grid consistent with the overall grid for each window, and calculate the window amplitude spectrum; Perform peak identification, energy calculation, and percentage calculation in each window; The primary resonance frequency is acquired in each window and formed into a primary resonance frequency sequence in chronological order.

[0012] Optionally, the step of selecting the principal resonance frequency of the first window as a reference, calculating the drift of the principal resonance frequency of each window, and generating a relative drift ratio sequence specifically includes: Select the main resonant frequency of the first window as the reference main frequency; Calculate the drift relative to the reference frequency for each of the second to the last windows; Calculate the arithmetic mean of the main resonance frequencies of the entire sequence, and terminate the process when the mean is zero; The relative drift ratio sequence is obtained by calculating the ratio of the drift amount of each window to the average value of the main resonant frequency.

[0013] Optionally, the step of calculating the ratio of the relative drift difference to the drift amount combination of two adjacent windows to obtain the main frequency stability of each window and calculating the mean stability specifically includes: Calculate the relative drift difference for any two adjacent windows; Set a non-zero constant related to the number of windows as the limiting parameter; A dimensionless ratio is constructed based on the relative drift difference and the sum of the drift amounts of the two windows, and the limiting process is performed according to the limiting parameters. The frequency stability of each window can be obtained by subtracting the dimensionless ratio from one. The arithmetic mean of the stationarity of the dominant frequency from the third window to the last window is used to obtain the mean stationarity value.

[0014] Optionally, the calculation of the average absolute value of the relative drift ratio, combined with the mean stability, constructs a health index, and classifies the data according to a threshold, outputting the main resonance frequency, drift amount, relative drift ratio, main frequency stability, mean stability, average absolute drift value, and health index. Specifically, this includes: Calculate the average absolute value of the relative drift ratio sequence; A health index is constructed based on the mean and absolute value of stability. The structure is classified according to the following thresholds: a health index greater than 0.8 indicates structural stability, a health index between 0.5 and 0.8 indicates metastable condition, and a health index less than 0.5 indicates abnormal vibration. Output a set of results including the main resonance frequency sequence, drift sequence, relative drift ratio sequence, main frequency stationarity sequence, mean stationarity, mean absolute value, and health index.

[0015] The present invention has the following beneficial effects: 1. After the sensors are installed at equal intervals along the pump valve axis, the zero-bias parameters of each accelerometer are obtained through multi-point sampling under static, pressureless conditions. During operation, all raw signals are calibrated point-by-point. Unlike existing technologies that uniformly set the zero bias to a constant or rely on manual adjustment, this method obtains the zero bias through automatic static averaging, which better reflects the field environment and individual sensor differences, solving the data distortion problem caused by inherent deviations between different sensors within the same equipment. On the one hand, the averaging calculation of static samples ensures stable zero-bias parameters; on the other hand, online calibration of operating signals can promptly counteract interference in the early stages of equipment vibration.

[0016] 2. This scheme, based on all corrected acceleration signals, sets a uniform sampling period and total duration, and performs an arithmetic average of multi-channel sensor data at each sampling time to form a global synthetic vibration sequence. Unlike existing systems that commonly use independent channel-by-channel monitoring or simple weighted fusion, this scheme adopts an equal-weighted arithmetic average method to unbiasedly fuse multi-point information. This avoids dependence on single-point faults and weakens the anomalies caused by individual sensor failures, thereby improving the overall system's ability to capture structural vibration characteristics. Simultaneously, setting a unified clock for synchronous sampling ensures the consistency of cross-channel signals, avoiding phase misalignment and spectral artifacts caused by asynchronous clocks. This fusion scheme has a significant effect on achieving real-time online monitoring and rapid early warning, enabling timely capture of subtle vibration changes when pumps and valves show early damage. Furthermore, it is more economical and feasible in terms of hardware deployment costs than introducing complex weighting algorithms or additional sensor redundancy.

[0017] 3. This scheme calculates the sampling frequency based on a unified sampling period and constructs a discrete frequency set from zero to the Nyquist frequency. At each frequency point, the superimposed amplitude of the sine and cosine projections of the synthesized vibration sequence is calculated to form a complete amplitude spectrum. Existing technologies often use Fast Fourier Transform to obtain the spectrum, but this scheme abandons the coarse-grained operation of converting the entire sequence at once, instead performing customized calculations for each discrete frequency point. Therefore, it can accurately calculate the energy distribution at local frequencies with limited data, which is particularly helpful in capturing narrow band and narrow peak information. Furthermore, by explicitly calculating the amplitude independently for each frequency point, this scheme has higher frequency resolution and noise resistance, and can distinguish subtle peak shifts caused by adjacent operating conditions or small structural changes in pump and valve vibration signals. This approach balances frequency domain analysis accuracy and online computational cost, making it suitable for resource-constrained edge computing scenarios.

[0018] 4. Based on the complete amplitude spectrum, this scheme performs neighborhood peak determination on all comparable interior points to form a candidate peak set; then, the energy of each peak is calculated and its energy percentage is obtained, and finally, the peak with the largest percentage and the smallest frequency value is selected as the main resonance frequency. Unlike traditional techniques that only select the point with the largest amplitude, this scheme considers both energy distribution and frequency priority, avoiding peak misjudgment caused by occasional noise or harmonic interference; in the case of parallel peaks, the smaller frequency value is selected, effectively improving the accuracy of primary resonance mode identification. This innovation not only simplifies the main peak determination process in multi-peak environments, but also enhances the method's ability to distinguish near-mode frequencies, making it particularly suitable for multi-resonance systems such as pumps and valves. In practical applications, accurate identification of the main resonance frequency is crucial for subsequent drift analysis and health assessment. This scheme reduces errors caused by peak drift or subharmonics, providing a reliable starting point for the entire assessment system.

[0019] 5. This scheme introduces a sliding time window and fixed step size strategy to perform segmented analysis of the synthesized vibration sequence. By calculating the frequency grid and amplitude spectrum of each window and repeatedly performing peak identification and dominant frequency determination in each window, a continuous dominant resonance frequency sequence can be formed in the time dimension. Compared with static spectrum analysis that treats the entire signal as a whole, the sliding window method can reveal the dynamic evolution trend of the dominant frequency with operating time, so that the monitoring of pump and valve operating status is no longer limited to the spectral characteristics at a single moment, but extended to the change trajectory of the time series. This is of great significance for capturing frequency drift caused by temperature changes, load fluctuations, or microcrack propagation. At the same time, this multi-window processing can achieve efficient online monitoring by concurrently operating on parallel windows while ensuring real-time performance, meeting the needs of continuous health assessment in industrial sites.

[0020] 6. This scheme uses the main resonant frequency of the first window as a reference, calculates the difference between the subsequent windows and the reference frequency, and then performs a ratio calculation with the average main frequency of the entire sequence to construct a relative drift ratio sequence. Using both the reference frequency and the average frequency as dual references, compared to methods that only calculate the absolute frequency difference, it better reflects the degree of deviation of structural changes relative to the overall operating level, eliminating the overall frequency shift caused by differences in operating conditions or ambient temperature drift, thus focusing on local frequency shifts caused by equipment damage or fatigue. Through the relative drift ratio sequence, more sensitive identification can be achieved in the early stages of minor damage, and interference from environmental noise and baseline drift can be reduced.

[0021] 7. After obtaining the relative drift ratio sequence, this scheme calculates the drift difference between adjacent windows and constructs a dimensionless stationarity function by combining the sum of the drift amounts of the two windows. The stationarity value for each window is obtained through subtraction and amplitude limiting operations, and the overall average is taken to obtain the final mean stationarity value. This method innovatively utilizes the ratio of the relative drift difference to the sum of the overall drift amounts to achieve a joint measurement of the rate and magnitude of frequency change; the amplitude limiting parameter ensures that the stationarity value is within a reasonable range, avoiding instability judgments caused by extreme fluctuations. Compared with methods that use only a single indicator to measure frequency fluctuations, the stationarity function is more sensitive to identifying sudden shocks or transient faults, while filtering out interference from small fluctuations during normal operation. This stationarity index provides an important dimension for subsequent health indices, enabling the overall evaluation system to consider both the amplitude of frequency drift and the continuity and stability of the drift process.

[0022] 8. The final step combines the mean stability and the average absolute value of the relative drift ratio to generate a health index through simple arithmetic. Based on preset thresholds, the structural state is categorized into three levels: "structurally stable," "metastable," and "abnormal vibration." Compared to complex machine learning classification models, this index construction requires only two indicators to quantify the structural state, reducing reliance on extensive historical data labeling and model training, and avoiding problems such as model overfitting and poor interpretability. The three-level grading method aligns closely with industrial maintenance practices, enabling maintenance personnel to quickly understand the meaning of the health index and take appropriate measures. Simultaneously, the simple arithmetic calculation of multiple indicators provides online real-time computation capabilities, meeting the needs for rapid on-site early warning and periodic diagnosis. This innovation achieves efficient and visualized assessment of pump and valve structural health while maintaining a lightweight system. Attached Figure Description

[0023] Figure 1 This is a schematic diagram of the process of the present invention. Detailed Implementation

[0024] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0025] Example, refer to Figure 1 A mechanical sensor-driven method for assessing the health of pumps and valves, comprising: Multiple acceleration sensors are axially installed on the pump and valve housing. Static sampling is performed, and the static average value of each sensor is calculated. Based on the static average value, zero-bias correction is performed on the running sampling sequence to obtain the correction sequence. Set a uniform sampling period and total duration, synchronously collect the calibration sequences of each sensor based on a uniform clock, perform an arithmetic average at each sampling moment, and generate a synthetic vibration sequence; Establish discrete frequency coordinates, and calculate the superimposed amplitude of sine and cosine waves around each discrete frequency of the synthesized vibration sequence to form an amplitude spectrum; Perform neighborhood comparisons within the amplitude spectrum to obtain a set of peaks, calculate the energy and energy percentage of each peak, and determine the main resonance frequency based on the peak with the largest energy percentage and the smallest frequency value. Set the sliding time window and step size, extract the window signals in sequence and calculate the window amplitude spectrum to obtain the main resonance frequency of each window; The main resonance frequency of the first window is selected as the benchmark, and the drift of the main resonance frequency of each window is calculated to generate a relative drift ratio sequence. Calculate the ratio of the relative drift difference to the drift amount combination between two adjacent windows to obtain the main frequency stationarity of each window and calculate the mean stationarity. Calculate the average absolute value of the relative drift ratio, construct a health index by combining it with the mean of the stability, complete the classification based on the threshold, and output the main resonance frequency, drift amount, relative drift ratio, main frequency stability, mean stability, average absolute value of drift, and health index.

[0026] First, multiple accelerometers are axially mounted on the pump and valve housing and static sampling is performed. After calculating the static average value, the runtime data is zero-biased corrected, which solves the measurement error caused by the initial offset of the sensors and improves the accuracy and consistency of subsequent vibration signals. Next, a unified sampling period and duration are set and samples are collected synchronously. The multi-channel corrected data are fused by arithmetic averaging, eliminating the interference of single-point mounting differences and occasional noise on the overall signal. This fusion method balances real-time performance and robustness, which is beneficial for online monitoring applications. Subsequently, by constructing discrete frequency coordinates and calculating the sine and cosine projection amplitudes at each frequency point, the limitations of traditional one-time fast Fourier transform in terms of resolution and noise resistance are overcome, and the local frequency band energy distribution can be accurately obtained, providing a high-precision spectrum for downstream peak identification. Then, spectral peaks are identified by neighborhood comparison, and the main resonant frequency is selected based on energy proportion and frequency priority, perfectly avoiding harmonic interference or noise peak misjudgment and ensuring that the selected frequency corresponds to the true structural natural vibration mode. The signal is further divided using a sliding time window, and the dominant frequency of each window is extracted, extending static dominant frequency identification to time-series analysis, making the method more comprehensive in its grasp of vibration drift trends. Then, the drift is calculated by comparing the difference between the dominant frequency of the reference window and the dominant frequencies of each window, and then ratiomed to the average dominant frequency. This successfully eliminates the overall frequency band shift caused by changes in operating conditions or ambient temperature, improving the sensitivity to minor equipment damage. Furthermore, a stability function is constructed using the ratio of the drift difference between adjacent windows to the drift amount, and the average is taken to obtain the mean stability value. This provides a quantitative indicator for vibration stability that reflects both the rate of change and filters out normal fluctuations. Finally, a health index is constructed by combining the average absolute value of the relative drift ratio and the mean stability value, and output in a graded manner. This compresses complex vibration characteristics into easily understandable health grading results, facilitating maintenance personnel to quickly determine the structural condition and take appropriate measures.

[0027] The process involves axially mounting multiple acceleration sensors on the pump / valve housing, performing static sampling, calculating the static average value of each sensor, and applying zero-bias correction to the operational sampling sequence based on the static average value to obtain a correction sequence. Specifically, this includes: Select multiple installation points at equal intervals along the axis of the pump and valve housing and number them. Install piezoelectric accelerometers and record the correspondence between the installation points and the sensor numbers. Under conditions where the equipment is stationary and pressureless, perform static sampling of all sensors for a uniform duration to obtain the static sampling sequence of each sensor; The arithmetic mean of the static acceleration for each sensor is calculated as the zero bias parameter; The original acceleration sequence was collected under the operating conditions of the equipment, and each time sample was calibrated according to the zero bias parameter to obtain the calibrated acceleration sequence of each sensor.

[0028] Further specific implementation steps include: Select equal intervals along the axial direction on the surface of the pump / valve housing under test. There are 1 installation point, each numbered as follows: The physical location of each point is denoted as ,exist Piezoelectric accelerometers are installed on each of them. ;in, The total number of installed accelerometer sensors; Number the sensors; For the first The physical location of each mounting point on the pump / valve housing; For installation in location The No. 1 acceleration sensor; Activate each sensor when the equipment is stationary and pressure-free. Perform sampling and record One time series sample: , ;in, The number of statically calibrated samples; For time indexing; For the first The first accelerometer sensor in a stationary state Acceleration readings at each sampling point; Calculate the first Average static acceleration of sensor number ; Real-time acceleration signals are collected while the equipment is in operation. And perform static correction: ;in, For the first Sensor number 1 is in operation. The original acceleration of each sampling point; For the first The first sensor Corrected acceleration at each sampling point.

[0029] First, installation points are selected and numbered at equal intervals along the pump and valve housing axis to ensure a one-to-one correspondence between the physical location of the sensors and subsequent data; this engineered layout avoids blind spots caused by random placement. Second, under static and pressureless conditions, all sensors are simultaneously subjected to static sampling for a uniform duration to obtain the static acceleration sequence of each sensor; this process ensures that the calculated zero-bias parameter reflects the true static state. Subsequently, the static average value is calculated for each sensor as the zero-bias parameter, and the corresponding zero bias is subtracted point by point from the original acceleration data during operation; compared with traditional fixed zero-bias values ​​or manual zeroing methods, the zero-bias parameter obtained by this solution through fully automatic static calibration is more repeatable and adaptable to the field. Through this series of steps, static offsets caused by sensor manufacturing errors, installation angles, or environmental influences are eliminated, improving the inherent consistency of multi-channel vibration data and laying a precise foundation for subsequent vibration sequence fusion and analysis. In addition, the step-by-step calibration strategy reduces manual debugging costs, minimizes human intervention at the maintenance site, and also gives the system higher adaptability and scalability under complex operating conditions.

[0030] The process of setting a unified sampling period and total duration, synchronously acquiring the calibration sequences of each sensor based on a unified clock, and performing an arithmetic mean at each sampling moment to generate a synthetic vibration sequence specifically includes: Set a uniform sampling period and total sampling duration, and calculate the number of samples to be collected; The discrete time series of calibrated acceleration is obtained for each sensor according to a unified clock. At each discrete time step, the calibration values ​​of all sensors are arithmetically averaged to generate a synthetic vibration sequence.

[0031] Further specific implementation steps include: Set a uniform sampling period The total sampling time is To obtain the total number of samples ; For each sensor This forms a vibration time series: ;in, ; For the first The calibrated acceleration timing of sensor number 1; Constructing a global synthesized signal: ;in, This represents the average response of all sensors.

[0032] The challenges of multi-channel data synchronization and fusion are addressed through standardized sampling configurations and unbiased fusion algorithms. First, a unified sampling period and total duration are set, and the total number of samples is calculated, ensuring the uniformity and integrity of data coverage throughout the evaluation period. Second, acceleration sequences from all calibrated sensors are synchronously acquired using a unified clock, avoiding phase misalignment and spectral artifacts caused by inconsistent channel timing; this is particularly important under high-speed or rapid-impact conditions. Then, an arithmetic mean is performed on the multi-channel data at each discrete moment to generate a globally synthesized vibration sequence. This weighted averaging method effectively reduces the abnormal influence of individual faulty sensors while enhancing the representativeness of the overall structural vibration characteristics. Unlike the weighted fusion or principal component extraction commonly used in existing technologies, this strategy ensures low-cost real-time computation through a simple arithmetic mean and, to some extent, mitigates sporadic failures of individual sensors. The resulting synthesized sequence, processed by this method, retains the key characteristics of pump and valve vibration while suppressing random noise and occasional impacts, improving the accuracy and stability of frequency domain analysis.

[0033] The process of establishing discrete frequency coordinates and calculating the superimposed sine and cosine amplitudes of the synthesized vibration sequence around each discrete frequency to form an amplitude spectrum specifically includes: The sampling frequency is calculated based on the sampling period, and the set of discrete frequencies up to the Nyquist frequency is determined. For each frequency in the discrete frequency set, the superposition intensity of the cosine component and the superposition intensity of the sine component are calculated for the synthesized vibration sequence, and the amplitude of the corresponding frequency point is obtained based on the combination of the two. The output covers the amplitude spectrum of all discrete frequencies.

[0034] Further specific implementation steps include: Set the sampling frequency to ; Construct the first Discrete frequency points: , , ;in, For discrete frequency indexing; Indexed by maximum frequency; For the first A discrete frequency; Constructing the frequency domain magnitude function is as follows: ;in, For the amplitude spectrum at frequency The amplitude at that point will be the time-domain signal. Projection to frequency The amplitude is synthesized by combining the cosine and sine of the base.

[0035] By calculating the superposition amplitude of cosine and sine components at discrete frequency points one by one, this method overcomes the performance trade-off between online computation and frequency resolution inherent in traditional Fast Fourier Transform (FFT). First, the sampling frequency is calculated based on a uniform sampling period, generating a discrete frequency set covering up to the Nyquist frequency, ensuring a complete scan of possible resonant modes. This process differs from fixed frequency band segmentation or coarse-grained frequency band division, improving the precision of frequency coverage. Subsequently, the cosine and sine projection amplitudes of the synthesized vibration sequence are calculated at each discrete frequency point and synthesized to obtain the amplitude. This accurately reflects the local energy distribution while avoiding the high latency and difficulty in online updates associated with one-time conversion of the entire sequence. This frequency-point-by-frequency calculation strategy maintains high frequency resolution even with limited signal length or frequent window switching, and is particularly effective for capturing narrow spectral peaks in pumps and valves caused by the propagation of minute cracks or frictional changes. Furthermore, this method achieves frequency domain analysis with controllable computational load, making it suitable for resource-constrained field equipment. Compared with existing FFT-based batch spectrum generation, this method not only improves the ability to identify narrowband frequency variations, but also simplifies the complexity of online implementation, laying a high-reliability frequency domain feature foundation for subsequent peak identification.

[0036] The process of performing neighborhood comparisons within the amplitude spectrum to obtain a peak set, calculating the energy and energy percentage of each peak, and determining the main resonance frequency based on the peak with the largest energy percentage and the smallest frequency value specifically includes: Perform left and right neighborhood amplitude comparisons at comparable in-place points of the amplitude spectrum to generate a set of candidate peaks; For each peak in the candidate peak set, calculate the peak energy by squared amplitude, and sum the energy across the entire frequency band; Peak identification is terminated when the energy across the entire frequency band is zero or the candidate peak set is empty. Calculate the energy percentage of each peak; The main resonance frequency is determined by the peak with the largest energy percentage. When the energy percentages are equal, the frequency corresponding to the peak with the smaller frequency value is selected.

[0037] Further specific implementation steps include: Extract all local maxima points, and the decision criterion is performed on interior points that are comparable to their left and right neighbors: , The frequency indices that meet the conditions are collected as follows: And construct a peak frequency set: ;in, For the first The frequency index value that was determined to be a local maximum; The peak number; The number of peaks identified; Includes the frequency values ​​corresponding to all identified peaks. ; For the first The frequency corresponding to each peak; Calculate the first Energy values ​​of each peak: , ;in, For the first The spectral energy of each peak; Calculate the total energy across the entire frequency band ; like If no valid vibration data is output, the sensor connection, operating conditions, and sampling parameters need to be checked, and the process should be stopped. like If no valid vibration peak is output, the sensor connection, operating conditions, and sampling parameters need to be checked, and the process should be stopped. Otherwise, calculate the first Energy percentage of each peak ; The frequency with the largest energy proportion is selected as the main resonant frequency. If there are two peaks with the same maximum energy proportion, the peak with the smallest frequency value is selected as the main peak. Specifically: make , , ;in, For index set; Returns the set of indices that maximize the target; The peak index corresponding to the largest energy percentage; This is the global primary resonance frequency.

[0038] First, a neighborhood amplitude comparison is performed at comparable interior points, selecting only peak candidates where both left and right neighboring points are smaller than the current point. This ensures the selection of true local maxima. Compared to the global amplitude threshold method, this neighborhood comparison is adaptive, maintaining robust peak identification even with varying noise levels. Next, the energy of each peak in the candidate peak set is calculated based on the square of its amplitude, and the total energy across the entire frequency band is summed. The energy percentage is used as the criterion for determining the dominant peak, completely avoiding the drawback of relying solely on the highest amplitude, which is susceptible to occasional interference. Furthermore, in cases of equal percentages, the lower frequency is prioritized, focusing on the first resonance mode, aligning with the focus on primary resonance in pump and valve vibration monitoring. This method not only distinguishes between aliased peaks, harmonic peaks, and noise peaks but also quantifies the relative importance of each peak in the overall vibration energy distribution through energy percentage. Compared to traditional peak extraction based on amplitude ranking, this method offers higher accuracy in identifying structural multiharmonic phenomena, improving the reliability of subsequent drift and health assessments, while reducing false alarms and false negatives, providing a solid dominant frequency source for pump and valve fault early warning.

[0039] The process of setting a sliding time window and step size, extracting window signals sequentially, calculating the window amplitude spectrum, and obtaining the main resonance frequency of each window specifically includes: Set the sliding time window length and step size, convert them to integer sample numbers, and calculate the number of windows; Calculate the start and end indices of each window according to its sequence number, and extract the window signal; Establish a discrete frequency grid consistent with the overall grid for each window, and calculate the window amplitude spectrum; Perform peak identification, energy calculation, and percentage calculation in each window; The primary resonance frequency is acquired in each window and formed into a primary resonance frequency sequence in chronological order.

[0040] Further specific implementation steps include: Set the length of each sliding time window to... The step size between every two windows is ; Set the integer sample length and step size as follows: , ;in, The number of samples within a single time window; The sample length at the start of the time window; And calculate the total number of windows. ; Set the starting sample index for each window as follows: , ;in, Number the windows; For the first The starting discrete index of each window in the entire sequence; Extract the first One window signal: , ;in, For window Inner The signal value of each sample; set up , , ;in, For window Maximum frequency index within; For window Inner One dispersion frequency; Constructing the window magnitude function: ;in, For window The amplitude function within; Calculate the total energy within the window: ;in, For window Total spectral energy; Peak determination within the window is as follows: right , The indexes that meet the conditions will be combined. Construct peak frequency set ;in, For window Inner Frequency index of each peak; For window The number of peaks; For window Set of internal peak frequencies; like or Output the first If no valid vibration data is found for the window, the sensor connection, operating conditions, and sampling parameters need to be checked, and the process should be stopped. Otherwise, calculate: , ;in, , windows respectively Inner The energy and energy percentage of each peak; The resonant frequency within the window is obtained as follows: make , , ;in, For window The set of peak indices; For window The strongest peak index; For window The main resonant frequency.

[0041] A sliding time window partitioning and multi-window dominant frequency sequence extraction strategy is proposed to compensate for the limitations of static spectrum from a dynamic monitoring perspective. First, the sliding window length and step size are set, and the time window is mapped to the corresponding number of samples and the total number of windows, ensuring the controllability of continuous signal segmentation. Then, the signal of each window is extracted sequentially, and the window amplitude spectrum is calculated based on a frequency grid consistent with the overall signal. The aforementioned peak identification and dominant frequency determination steps are repeated, thereby forming a dominant resonance frequency sequence in the time dimension. Unlike analyzing only the entire signal segment or non-overlapping window analysis, the sliding overlapping window can more finely characterize the frequency change trend with operating conditions, balancing time resolution and spectral stability. During the operation of pump and valve equipment, different loads, speeds, or temperature changes can cause dominant frequency shifts. Sliding window processing can capture these changes in real time, providing time-series-level dynamic information for operational status diagnosis. This method has significant advantages in detecting early, minor damage: even if the damage only causes spectral shifts within a short time window, it can be marked and fed back through the overlapping strategy. In addition, windowed processing supports parallel computing, which can meet the real-time requirements of on-site online monitoring and improve the sensitivity of pump and valve health assessment and continuous monitoring capabilities.

[0042] The step of selecting the main resonance frequency of the first window as a benchmark, calculating the drift of the main resonance frequency of each window, and generating a relative drift ratio sequence specifically includes: Select the main resonant frequency of the first window as the reference main frequency; Calculate the drift relative to the reference frequency for each of the second to the last windows; Calculate the arithmetic mean of the main resonance frequencies of the entire sequence, and terminate the process when the mean is zero; The relative drift ratio sequence is obtained by calculating the ratio of the drift amount of each window to the average value of the main resonant frequency.

[0043] Further specific implementation steps include: Using the first window's main frequency as a baseline, calculate the drift amount for each window: , ;in, For window The frequency shift of the main frequency relative to window 1; Calculate the average clock frequency of the entire sequence ; like The average output frequency is zero, and frequency normalization is invalid. The operating conditions and sampling configuration need to be checked, and the process should be stopped. Constructing the normalized relative drift ratio: , ;in, For window The relative frequency drift ratio.

[0044] By employing a drift normalization strategy based on both a reference window and an average window, the overall frequency shift caused by environmental temperature and load variations is effectively isolated. First, using the dominant resonant frequency of the first window as a reference, the drift of each subsequent window relative to the reference frequency is calculated, directly reflecting the relative frequency change in the initial state. This approach more intuitively presents the overall frequency shift trend caused by damage than simply calculating the difference between consecutive windows. Then, the arithmetic mean of the dominant frequencies of the entire sequence is calculated as a normalization factor, and the drift of each window is ratioized to the average dominant frequency, resulting in a dimensionless relative drift ratio. This normalization ensures that the results are unaffected by differences in the absolute frequency baseline of the equipment, making them more comparable across equipment. Compared to existing methods that rely solely on absolute drift or drift relative to the previous window, this approach considers both the initial state and the overall average level, reducing errors caused by initial calibration deviations or short-term abnormal fluctuations between windows, and improving the stability and reliability of the drift sequence. The relative drift ratio sequence generated by this method intuitively reflects the deviation trend of the dominant frequency during equipment operation, providing more accurate and comparable input data for subsequent stability calculations and health index construction.

[0045] The calculation of the ratio of the relative drift difference to the drift amount combination between two adjacent windows, obtaining the main frequency stability of each window, and calculating the mean stability specifically includes: Calculate the relative drift difference for any two adjacent windows; Set a non-zero constant related to the number of windows as the limiting parameter; A dimensionless ratio is constructed based on the relative drift difference and the sum of the drift amounts of the two windows, and the limiting process is performed according to the limiting parameters. The frequency stability of each window can be obtained by subtracting the dimensionless ratio from one. The arithmetic mean of the stationarity of the dominant frequency from the third window to the last window is used to obtain the mean stationarity value.

[0046] Further specific implementation steps include: Calculate the relative drift change between any two adjacent windows: , ;in, For window With window The difference in adjacent relative drift; Set the zero constant for prevention to be... ; The stationarity function is constructed as follows: , ;in, The value is related to the frequency stability; the closer the value is to 1, the more stable the frequency is between adjacent windows. Calculate the mean stationarity .

[0047] By constructing a dimensionless stationarity index using the drift difference between two adjacent windows and the sum of their drift values, the problem that drift values ​​alone cannot reflect the rate of change and stability is solved. Specifically, firstly, the relative drift difference between any two adjacent windows is calculated to quantify the degree of abrupt change in the dominant frequency along the time axis. Then, a non-zero limiting constant related to the number of windows is set to prevent the denominator from being zero and to control extreme values, ensuring that the stationarity function fluctuates within a controllable range. Next, the drift difference is combined with the sum of the drift values ​​of the two windows, and the stationarity value is obtained by subtracting this ratio from one. The closer the value is to one, the more stable the frequency change. This formula takes into account both the drift amplitude and the rate of change, reflecting both the sharp drop in stationarity caused by sudden jumps and maintaining a high stationarity value during small, normal fluctuations. Finally, the stationarity of the third window and subsequent windows is averaged to obtain the overall mean stationarity value, providing a quantitative measure of the continuity and stability of the dominant frequency drift process for health assessment. Compared with traditional methods that only consider drift amount or drift rate, this stability index is more comprehensive: it can capture frequency jumps and smooth normal fluctuations, improving the sensitivity of identifying early faults or structural loosening, while avoiding misjudging small vibration changes under normal operating conditions.

[0048] The calculation of the average absolute value of the relative drift ratio, combined with the mean stability, constructs a health index. Based on thresholds, it performs classification and outputs the main resonant frequency, drift amount, relative drift ratio, main frequency stability, mean stability, average absolute drift value, and health index. Specifically, this includes: Calculate the average absolute value of the relative drift ratio sequence; A health index is constructed based on the mean and absolute value of stability. The structure is classified according to the following thresholds: a health index greater than 0.8 indicates structural stability, a health index between 0.5 and 0.8 indicates metastable condition, and a health index less than 0.5 indicates abnormal vibration. Output a set of results including the main resonance frequency sequence, drift sequence, relative drift ratio sequence, main frequency stationarity sequence, mean stationarity, mean absolute value, and health index.

[0049] Further specific implementation steps include: Calculate the absolute value of the average drift ; The health index of the pump and valve structure is constructed as follows: ; Classified according to health index: like If so, the structure is considered stable and in good condition; like If so, it is determined to be structurally metastable, and observation is recommended; like If so, it is determined to be abnormal structural vibration, and maintenance is recommended; Output result set .

[0050] By constructing a health index by combining the average absolute value of the relative drift ratio and the mean stability, and classifying it according to thresholds, a direct quantification and classification of the pump and valve structural status is achieved. First, the average absolute value of the relative drift ratio sequence is calculated, reflecting the average intensity of the main frequency deviation from the overall level. Then, this value is combined with the overall stability mean using simple arithmetic to generate the health index, ensuring that the result simultaneously considers the amplitude of frequency shift and the stability of frequency changes. This index formula is simple in structure, highly interpretable, and does not rely on complex models or large amounts of historical data, reducing deployment and maintenance costs. Finally, three threshold levels are set to map the health index to "structural stability," "structural metastable," and "vibration anomaly," closely aligning with industrial maintenance needs. This allows maintenance personnel to quickly determine the structural condition and execute corresponding maintenance strategies based on the returned results. Compared to fault diagnosis methods that rely on machine learning or empirical rules, this method provides a lightweight, real-time, and highly interpretable health assessment tool, not only improving early fault warning capabilities but also enhancing the transparency and reliability of maintenance decisions.

[0051] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0052] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for assessing the health of a pump valve driven by a mechanical sensor, characterized in that, include: Multiple acceleration sensors are axially installed on the pump and valve housing. Static sampling is performed, and the static average value of each sensor is calculated. Based on the static average value, zero-bias correction is performed on the running sampling sequence to obtain the correction sequence. Set a uniform sampling period and total duration, synchronously collect the calibration sequences of each sensor based on a uniform clock, perform an arithmetic average at each sampling moment, and generate a synthetic vibration sequence; Establish discrete frequency coordinates, and calculate the superimposed amplitude of sine and cosine waves around each discrete frequency of the synthesized vibration sequence to form an amplitude spectrum; Perform neighborhood comparisons within the amplitude spectrum to obtain a set of peaks, calculate the energy and energy percentage of each peak, and determine the main resonance frequency based on the peak with the largest energy percentage and the smallest frequency value. Set the sliding time window and step size, extract the window signals in sequence and calculate the window amplitude spectrum to obtain the main resonance frequency of each window; The main resonance frequency of the first window is selected as the benchmark, and the drift of the main resonance frequency of each window is calculated to generate a relative drift ratio sequence. Calculate the ratio of the relative drift difference to the drift amount combination between two adjacent windows to obtain the main frequency stationarity of each window and calculate the mean stationarity. Calculate the average absolute value of the relative drift ratio, construct a health index by combining it with the mean of the stability, complete the classification based on the threshold, and output the main resonance frequency, drift amount, relative drift ratio, main frequency stability, mean stability, average absolute value of drift, and health index.

2. The method for assessing pump and valve health driven by a mechanical sensor according to claim 1, characterized in that, The process involves axially mounting multiple acceleration sensors on the pump / valve housing, performing static sampling, calculating the static average value of each sensor, and applying zero-bias correction to the operational sampling sequence based on the static average value to obtain a correction sequence. Specifically, this includes: Select multiple installation points at equal intervals along the axis of the pump and valve housing and number them. Install piezoelectric accelerometers and record the correspondence between the installation points and the sensor numbers. Under conditions where the equipment is stationary and pressureless, perform static sampling of all sensors for a uniform duration to obtain the static sampling sequence of each sensor; The arithmetic mean of the static acceleration for each sensor is calculated as the zero bias parameter; The original acceleration sequence was collected under the operating conditions of the equipment, and each time sample was calibrated according to the zero bias parameter to obtain the calibrated acceleration sequence of each sensor.

3. The method for assessing pump and valve health driven by a mechanical sensor according to claim 2, characterized in that, The process of setting a unified sampling period and total duration, synchronously acquiring the calibration sequences of each sensor based on a unified clock, and performing an arithmetic mean at each sampling moment to generate a synthetic vibration sequence specifically includes: Set a uniform sampling period and total sampling duration, and calculate the number of samples to be collected; The discrete time series of calibrated acceleration is obtained for each sensor according to a unified clock. At each discrete time step, the calibration values ​​of all sensors are arithmetically averaged to generate a synthetic vibration sequence.

4. The method for assessing pump and valve health driven by a mechanical sensor according to claim 3, characterized in that, The process of establishing discrete frequency coordinates and calculating the superimposed sine and cosine amplitudes of the synthesized vibration sequence around each discrete frequency to form an amplitude spectrum specifically includes: The sampling frequency is calculated based on the sampling period, and the set of discrete frequencies up to the Nyquist frequency is determined. For each frequency in the discrete frequency set, the superposition intensity of the cosine component and the superposition intensity of the sine component are calculated for the synthesized vibration sequence, and the amplitude of the corresponding frequency point is obtained based on the combination of the two. The output covers the amplitude spectrum of all discrete frequencies.

5. The pump valve health assessment method driven by a mechanical sensor according to claim 4, characterized in that, The process of performing neighborhood comparisons within the amplitude spectrum to obtain a peak set, calculating the energy and energy percentage of each peak, and determining the main resonance frequency based on the peak with the largest energy percentage and the smallest frequency value specifically includes: Perform left and right neighborhood amplitude comparisons at comparable in-place points of the amplitude spectrum to generate a set of candidate peaks; For each peak in the candidate peak set, calculate the peak energy by squared amplitude, and sum the energy across the entire frequency band; Peak identification is terminated when the energy across the entire frequency band is zero or the candidate peak set is empty. Calculate the energy percentage of each peak; The main resonance frequency is determined by the peak with the largest energy percentage. When the energy percentages are equal, the frequency corresponding to the peak with the smaller frequency value is selected.

6. The method for assessing pump and valve health driven by a mechanical sensor according to claim 5, characterized in that, The process of setting a sliding time window and step size, extracting window signals sequentially, calculating the window amplitude spectrum, and obtaining the main resonance frequency of each window specifically includes: Set the sliding time window length and step size, convert them to integer sample numbers, and calculate the number of windows; Calculate the start and end indices of each window according to its sequence number, and extract the window signal; Establish a discrete frequency grid consistent with the overall grid for each window, and calculate the window amplitude spectrum; Perform peak identification, energy calculation, and percentage calculation in each window; The primary resonance frequency is acquired in each window and formed into a primary resonance frequency sequence in chronological order.

7. The method for assessing pump and valve health driven by a mechanical sensor according to claim 6, characterized in that, The step of selecting the main resonance frequency of the first window as a benchmark, calculating the drift of the main resonance frequency of each window, and generating a relative drift ratio sequence specifically includes: Select the main resonant frequency of the first window as the reference main frequency; Calculate the drift relative to the reference frequency for each of the second to the last windows; Calculate the arithmetic mean of the main resonance frequencies of the entire sequence, and terminate the process when the mean is zero; The relative drift ratio sequence is obtained by calculating the ratio of the drift amount of each window to the average value of the main resonant frequency.

8. The method for assessing pump and valve health driven by a mechanical sensor according to claim 7, characterized in that, The calculation of the ratio of the relative drift difference to the drift amount combination between two adjacent windows, obtaining the main frequency stability of each window, and calculating the mean stability specifically includes: Calculate the relative drift difference for any two adjacent windows; Set a non-zero constant related to the number of windows as the limiting parameter; A dimensionless ratio is constructed based on the relative drift difference and the sum of the drift amounts of the two windows, and the limiting process is performed according to the limiting parameters. The frequency stability of each window can be obtained by subtracting the dimensionless ratio from one. The arithmetic mean of the stationarity of the dominant frequency from the third window to the last window is used to obtain the mean stationarity value.

9. The method for assessing pump and valve health driven by a mechanical sensor according to claim 8, characterized in that, The calculation of the average absolute value of the relative drift ratio, combined with the mean stability, constructs a health index. Based on thresholds, it performs classification and outputs the main resonant frequency, drift amount, relative drift ratio, main frequency stability, mean stability, average absolute drift value, and health index. Specifically, this includes: Calculate the average absolute value of the relative drift ratio sequence; A health index is constructed based on the mean and absolute value of stability. The structure is classified according to the following thresholds: a health index greater than 0.8 indicates structural stability, a health index between 0.5 and 0.8 indicates metastable condition, and a health index less than 0.5 indicates abnormal vibration. Output a set of results including the main resonance frequency sequence, drift sequence, relative drift ratio sequence, main frequency stationarity sequence, mean stationarity, mean absolute value, and health index.

Images