A new type of water pump pressure controller

Abstract

The application discloses a novel water pump pressure controller, comprising a transient waveform and vector acquisition module, a preceding locking matching module, a load vector projection calculation module, a quadrant anchor adjustment module and a feature dictionary self-learning module; a transient waveform snapshot is obtained by collecting pressure second-order difference zero-crossing points, and a load vector is formed in combination with torque and rotating speed; a dynamic time warping and power factor angle alignment output locking mark is used; a projection sum and an angle deviation are obtained by selecting a reference pump orthogonal projection; a minimum pump fine adjustment is selected according to the deviation sequence of the quadrant anchor, and the process is repeated until full locking; and the waveform, frequency and phase features are stored for self-learning. The application is related to transient waveform and load, eliminates phase deviation, decouples deviation step adjustment, and realizes smooth convergence of multiple pumps.

Description

Technical Field

[0001] This invention relates to the field of controller technology, and more specifically to a novel water pump pressure controller. Background Technology

[0002] In constant-pressure water supply scenarios such as water supply, HVAC, and industrial circulation, where multiple pumps need to operate in parallel, the controller must adjust the pump speed and switch pumps on and off in real time according to changes in pipeline pressure to maintain stable pressure. Currently, common constant-pressure water supply controllers mainly rely on pressure sensors to detect pipeline pressure, adjust the inverter output frequency through algorithms such as PID, and sequentially activate or deactivate pumps when the pressure exceeds set upper and lower limits. This type of control can basically meet the requirements under conditions of stable water flow and slow pressure changes.

[0003] However, in actual pipe networks, pressure fluctuations with obvious transient characteristics frequently occur due to random fluctuations in user water usage, rapid opening and closing of valves, and the start and stop of water pumps. Existing controllers mostly perform low-pass filtering or simple smoothing of pressure signals, only extracting the pressure amplitude for comparison with a threshold, thus losing the details of the transient waveform of pressure fluctuations and making it difficult to distinguish between short-term water hammer impacts and actual load changes. This leads to repeated pump switching under critical operating conditions, causing system oscillations and hydraulic shocks, affecting equipment lifespan and water supply quality.

[0004] Existing technologies typically rely on a single pressure threshold or PID control. When faced with heavy water usage, start-stop cycles, or water hammer, they only respond to the current pressure, lacking the ability to distinguish transient characteristics. This can easily lead to overshoot or continuous oscillation, causing overpressure shutdowns or pipeline fatigue. For example, in high-rise secondary water supply systems, repeated start-stop cycles during nighttime and early morning switching shorten motor life and cause pressure fluctuations. Furthermore, using pressure as the sole input fails to differentiate between increased water usage and pressure drops caused by leakage, resulting in energy waste or response lag when using conservative frequency reduction. Over long-term operation, impeller wear causes performance curve shifts, and fixed parameters gradually become ineffective, requiring manual calibration. When multiple pumps are connected in parallel, each pump independently adjusts its inlet and outlet pressures without coordinated locking. This leads to water grabbing or idling during load fluctuations, resulting in low overall efficiency. For instance, in industrial park parallel water pump cooling systems, frequent switching makes it difficult to maintain optimal operating conditions. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of existing technologies and propose a novel water pump pressure controller.

[0006] The technical solution provided by this invention to solve the above problems is: a novel water pump pressure controller, comprising: The transient waveform and vector acquisition module is used to acquire real-time pressure values ​​and perform second-order differential, detect zero-crossing points and capture data to form transient waveform snapshots, acquire torque and speed to form load operating point vectors, and transmit the two to the advance locking matching module; The pre-lock matching module is used to receive transient waveform snapshots and load operating point vectors, call standard feature codes to perform dynamic time warping comparison distance values, acquire power factor angle phase alignment, output a lock success flag, and transmit the load operating point vector and lock success flag to the load vector projection calculation module. The load vector projection calculation module is used to receive the load working point vector and the locking success mark, select the reference pump to calculate the reference axis angle, perform orthogonal projection to obtain the projection deviation and angle deviation, and transmit the two to the wheel anchoring adjustment module. The wheel anchoring adjustment module is used to receive the projection deviation and angle deviation and start anchoring, sort the projection deviation to select the minimum pump fine-tuning lock, recalibrate the reference axis, sort the angle deviation to select the minimum pump fine-tuning lock, repeat until full lock, and transmit the transient waveform snapshot, target locking frequency and power factor angle feature combination to the feature dictionary self-learning module.

[0007] Preferably, it also includes a feature dictionary self-learning module, which is used to receive transient waveform snapshots, target locking frequency and power factor angle feature combinations and add them to the feature dictionary, and replace the earliest or low frequency record when the entries are full.

[0008] Preferably, the transient waveform snapshot includes a pressure change sequence and a critical time point, and the load operating point vector includes mechanical torque and rotational speed.

[0009] Preferably, the successful locking indicator includes a similarity metric and a phase synchronization value, and the projection deviation includes a projection distance and a deviation direction.

[0010] Preferably, the angle deviation includes the angle difference and the angle sign; the target locking frequency includes the locking frequency value and the frequency tolerance; the power factor angle feature combination includes the phase feature and the combination weight; and the feature dictionary includes the feature record table and the replacement strategy.

[0011] Preferably, the transient waveform and vector acquisition module includes: The pressure differential acquisition submodule acquires a real-time pressure value sequence with a pressure value sampling interval of 1 millisecond. The sequence contains 1000 sampling points. The difference between each pressure value and the pressure value at the previous moment is calculated to obtain a first-order difference sequence. Then, the difference between adjacent elements of the first-order difference sequence is calculated to obtain a second-order difference sequence. The second-order difference sequence contains 998 signed values, thus obtaining the pressure differential sequence. The segmentation generation submodule, based on the pressure differential sequence, compares the signs of two adjacent differential values ​​in the sequence. When the sign changes from positive to negative or from negative to positive, the position is marked as a zero-crossing point. 64 data points before and after each zero-crossing point are extracted to generate waveform segments. All waveform segments are combined into a dataset to obtain a transient waveform snapshot. The integrated output submodule collects torque and speed values, combines the torque and speed values ​​into a two-dimensional vector as the load operating point vector, and encapsulates the waveform data array and vector components into the same data packet based on the transient waveform snapshot and the load operating point vector to generate waveform load coupling data.

[0012] Preferably, the advance locking matching module includes: The waveform matching submodule receives transient waveform snapshots and load operating point vectors, aligns the transient waveform snapshots with the reference waveform in the standard feature code through dynamic time warping, calculates and accumulates the Euclidean distance point by point, takes the minimum accumulated distance value, and generates the waveform matching distance value. The phase alignment submodule extracts the power factor angle from the load operating point vector based on the waveform matching distance value and the load operating point vector. It then retrieves the corresponding standard power factor angle from the standard feature code according to the waveform matching distance value, calculates the difference between the power factor angle and the standard power factor angle, and obtains the phase offset. The lock determination submodule compares the waveform matching distance value with a preset distance threshold and the phase offset with a preset angle threshold based on the waveform matching distance value and phase offset. When both comparison results indicate a less than relationship, a lock success flag is output.

[0013] Preferably, the load vector projection calculation module includes: The axis angle calculation submodule extracts the reference pump number from the lock success identifier, retrieves the ordinate and abscissa components of the corresponding load working point vector in the preset reference pump data table based on the number, divides the ordinate component by the abscissa component and takes the arctangent to obtain the angle value as the reference axis angle, and generates the reference axis angle. The orthogonal calculation submodule calls the reference axis angle and the load working point vector, calculates the dot product of the load working point vector and the reference axis unit vector to obtain the projection scalar, multiplies the projection scalar by the reference axis unit vector to obtain the projection vector, calculates the Euclidean norm of the difference vector between the load working point vector and the projection vector as the projection deviation, and calculates the absolute value of the difference between the argument of the load working point vector and the reference axis angle as the angle deviation, thus obtaining the projection deviation and the angle deviation.

[0014] Preferably, the wheel anchoring adjustment module includes: The projection deviation sorting submodule receives the projection deviation and angle deviation, starts the anchoring program, iterates through the projection deviation values ​​of all pumps, compares the deviation values ​​of each pump pairwise, selects the pump with the smallest deviation value, performs fine-tuning locking on the pump and records its number, transmits the angle deviation, and obtains the locked pump number. The angle deviation locking submodule recalibrates the reference axis to point to the pump based on the locked pump serial number, iterates through the angle deviation values ​​of all pumps, compares the angle deviation values ​​of each pump pairwise, selects the pump with the smallest angle deviation value, performs fine-tuning locking on the pump and records its number, and obtains the reference locked pump number. The full-lock iterative loop submodule cyclically calls the projection deviation sorting and angle deviation locking steps. Each iteration is based on the set of unlocked pumps, and the locking status is updated using the locked pump number and the reference locked pump number. The above process is repeated until all pumps are locked. The transient waveform snapshot and the target locking frequency and power factor angle feature combination are combined to obtain the transient waveform snapshot, target locking frequency and power factor angle feature combination.

[0015] Preferably, the feature dictionary self-learning module includes: The boundary feature extraction submodule is used to perform normalization processing on the amplitude sequence of transient waveform snapshots, call the box dimension method to calculate the fractal dimension, call the Shannon entropy to calculate the information entropy, and take the mean after normalizing the fractal dimension and information entropy as the boundary representativeness. The transient waveform snapshot, boundary representativeness, target locking frequency and power factor angle features are passed to the feature fusion clustering module. The feature fusion and clustering submodule receives transient waveform snapshots, boundary representativeness, target locking frequency, and power factor angle features. Based on the second-order difference zero-crossing sequence of the transient waveform snapshot, it performs K-means clustering algorithm to obtain candidate clusters. Within each candidate cluster, it selects the entry with the maximum value based on the boundary representativeness as the boundary-retained entry. It calculates the geometric mean of the target locking frequency for the non-boundary-retained entries in the candidate cluster to obtain the fusion frequency. It calculates the vector median of the power factor angle features to obtain the fusion phase angle. It performs point-by-point median operation on the waveform amplitude of the non-boundary-retained entries to obtain the fusion waveform snapshot. It then transmits the fusion waveform snapshot, fusion frequency, fusion phase angle, and boundary-retained entry index to the fault trend monitoring module. The fault trend monitoring submodule receives fused waveform snapshots, fused frequencies, fused phase angles, and boundary retention entry indexes. It performs Welch's method on the fused waveform snapshots to calculate the power spectral density and extracts the power spectral centroid frequency. It then performs first-order difference calculations on the power spectral centroid frequency sequence and the fused phase angle sequence, respectively. It calculates the proportion of positive values ​​in the first-order difference of the power spectral centroid frequency and compares it with a threshold of 0.8. It determines whether the slope of the first-order difference of the fused phase angle remains positive and is greater than a preset threshold. When the proportion of positive values ​​continuously exceeds 0.8 and the phase angle slope remains positive, it triggers a cavitation warning signal. It accumulates the boundary representativeness of the triggered warning entries to the degradation cumulative integral value. When the degradation cumulative integral value exceeds a set threshold, it marks the corresponding boundary retention entry index as a degradation core feature and retains it in the feature dictionary. The retained degradation core feature entry index, along with the corresponding fused frequency, fused phase angle, and fused waveform snapshot, is transmitted to the original scheme's feature dictionary self-learning module for overwriting and replacement. The above scheme calculates the fractal dimension and information entropy of transient waveform snapshots through a boundary feature extraction module and normalizes them to obtain the boundary representativeness. This quantifies the importance of waveform features for each entry, avoiding the loss of entries with potential fault characterization capabilities when the original dictionary is simply replaced. The feature fusion clustering module clusters similar entries and selects boundary-retained entries based on the boundary representativeness. For non-boundary entries, it performs frequency geometric mean, phase angle vector median, and waveform amplitude point-by-point median calculations to generate fused waveform snapshots, fused frequencies, and fused phase angles, effectively compressing redundant data and maintaining the representative features within clusters. The fault trend monitoring module calculates the power spectrum centroid frequency of the fused waveform snapshots using the Welch method. Combined with the first-order difference statistics and slope judgment of the fused phase angle sequence, a cavitation warning signal is triggered when the dual trend conditions are met. The boundary representativeness of the warning entries is accumulated to the cumulative degradation integral value. Entries exceeding the threshold are marked as core degradation features and retained, enabling the dictionary to track the system degradation evolution over a long period and improving the proactive perception capability of abnormal operating conditions.

[0016] Compared with existing technologies, the advantages of this invention are: by processing the real-time pressure signal through second-order differential processing and capturing transient waveform snapshots at zero-crossing points, fluctuation characteristics are effectively preserved and noise is suppressed; the torque and speed are synchronously collected to form a load operating point vector, associating the dynamic waveform with the static load. Dynamic time warping is used to compare the distance between the waveform and the standard feature code, and the phase is aligned using the power factor angle to eliminate time scale distortion and phase offset, improving matching reliability. A reference pump is selected to calculate the reference axis angle, and the load vector is orthogonally projected to obtain the projection deviation and angular deviation, decoupling complex deviations. The projection deviation is sorted to select the pump with the smallest value for fine-tuning and locking. After recalibrating the reference axis, the angular deviation is sorted to select the pump with the smallest value for fine-tuning and locking, repeating until all are locked, achieving smooth convergence and avoiding oscillations. After locking, the waveform, frequency, and phase angle feature combination is stored in the dictionary. When the dictionary is full, the earliest or low-frequency record is replaced, allowing the system to accumulate typical operating condition characteristics. Subsequently, the repetitive pattern can be quickly locked, reducing computing resources and enhancing long-term adaptive capability. Attached Figure Description

[0017] The accompanying drawings, which are provided to further illustrate the invention and constitute a part of this invention, are illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an undue limitation of the invention.

[0018] Figure 1 This is a system block diagram of the present invention. Detailed Implementation

[0019] The following will describe in detail the implementation of the present invention with reference to the accompanying drawings and embodiments, so that the process of how the present invention uses technical means to solve technical problems and achieve technical effects can be fully understood and implemented accordingly.

[0020] In the description of this invention, it should be noted that the directional terms such as "center", "lateral", "longitudinal", "length", "width", "thickness", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", "clockwise", and "counterclockwise" indicate the orientation and positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. They should not be construed as limiting the specific protection scope of this invention.

[0021] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features. Thus, the use of "first" and "second" to define a feature may explicitly or implicitly include one or more of that feature, and in the description of this invention, "a number" means two or more, unless otherwise explicitly specified.

[0022] In this invention, unless otherwise explicitly specified and limited, the terms "assembly," "connection," and "joining" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can also refer to a mechanical connection; they can refer to a direct connection or a connection through an intermediate medium; or they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0023] It should be understood that, when used in this specification and the appended claims, the terms "comprising" and "including" indicate the presence of the described features, integrals, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.

[0024] It should also be understood that the terminology used in this specification of embodiments of the invention is for the purpose of describing particular embodiments only and is not intended to limit the embodiments of the invention. As used in this specification of embodiments of the invention and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0025] Specific embodiments of the present invention are shown in the accompanying drawings. A novel water pump pressure controller includes a transient waveform and vector acquisition module, a pre-locking matching module, a load vector projection calculation module, a wheel anchoring adjustment module, and a feature dictionary self-learning module. The transient waveform and vector acquisition module is used to acquire real-time pressure values ​​and perform second-order differential analysis, detect zero-crossing points and capture data as transient waveform snapshots, acquire torque and speed as load operating point vectors, and transmit both to the pre-lock matching module. The pre-lock matching module receives the transient waveform snapshot and load operating point vector, calls standard feature codes for dynamic time warping and distance comparison, acquires power factor angle phase alignment, outputs a lock success flag, and transmits the load operating point vector and lock success flag to the load vector projection calculation module. The load vector projection calculation module receives the load operating point vector and lock success flag, selects a reference pump to calculate the reference axis angle, performs orthogonal projection to obtain projection deviation and angle deviation, and transmits both to the wheel anchoring adjustment module. The wheel anchoring adjustment module receives the projection deviation and angle deviation and starts anchoring, sorts the projection deviation to select the pump with the smallest fine-tuning lock, re-calibrates the reference axis, sorts the angle deviation to select the pump with the smallest fine-tuning lock, repeats until full lock, and transmits the transient waveform snapshot, target locking frequency, and power factor angle feature combination to the feature dictionary self-learning module. The feature dictionary self-learning module receives transient waveform snapshots, target locking frequencies, and power factor angle feature combinations and adds them to the feature dictionary. When the dictionary is full, the oldest or lowest frequency record is replaced. The transient waveform snapshots include pressure change sequences and critical time points; the load operating point vector includes mechanical torque and rotational speed; the successful locking indicator includes similarity measurement and phase synchronization value; the projection deviation includes projection distance and deviation direction; the angle deviation includes angle difference and angle sign; the target locking frequency includes locking frequency value and frequency tolerance; the power factor angle feature combination includes phase features and combination weights; and the feature dictionary includes a feature record table and a replacement strategy.Among them, projection deviation refers to the error in distance and direction measured during the projection of the load vector, used to assess the degree of deviation between the current operating state and the ideal state; by calculating the projection of the load operating point on the reference axis, the pump's operating mode can be adjusted in a timely manner to improve efficiency; angular deviation refers to the angular error obtained when analyzing the load operating point, reflecting the difference in angle between the actual state and the desired state; this information can be used to further optimize the pump's operating parameters to achieve optimal performance; mechanical torque reflects the pump's rotational capacity during operation, usually calculated by the applied force and the distance between its point of application; rotational speed represents the rotational rate of the pump components, measured in revolutions per minute (RPM); similarity metric is used to determine the degree of similarity between transient waveform snapshots and standard feature codes, and is an important part of the matching process. Key parameters include: Phase synchronization value, which indicates the degree of alignment between the rotational speed and the random state of the transient waveform, affecting the dynamic response capability of the pump; Projection distance, which refers to the distance deviation measured when projecting from the ideal state under load, is an important indicator for evaluating pump performance; Deviation direction, which indicates the positive or negative direction of the projection deviation, provides directional guidance for adjusting the pump's operating state; Angle difference, which is the difference in angle between the actual operating state and the preset target state, reflects the degree of deviation of the pump from the calibration target; Angle sign, which indicates the positive or negative direction of the deviation, helps determine the direction of adjustment; Phase feature, which is a quantitative description of the power factor angle, affecting the system's power output to evaluate the equipment's working efficiency; and Combination weight, which assigns different importance to different features in multi-feature synthesis, affecting the allocation of feature weights and subsequent learning strategies.

[0026] For example, the transient waveform and vector acquisition module includes: The pressure differential acquisition submodule uses a multi-stage centrifugal pump with a rated head of 50 meters and a rated flow rate of 20 cubic meters per hour as an example. The pressure sensor is installed at the pump outlet flange, the sampling period is set to 1 millisecond, and 1000 sampling points are continuously collected to form a real-time pressure value sequence. For example, the pressure value of the first sampling point... The second one The third one The 4th The 5th The 6th The 7th The 8th The 9th The 10th Subsequent sampling data can similarly form a complete sequence. The difference between each pressure value and the pressure value at the previous moment is calculated to obtain a first-order difference sequence, the formula of which is: subscript Take values ​​from 2 to 1000. This is the first-order difference value, in megapascals (MPA). (Example:) , , Calculate point by point until 999 first-order difference values ​​are obtained. Then, calculate the difference between adjacent elements of the first-order difference sequence to obtain the second-order difference sequence. The formula is as follows: ,in The acceleration representing the rate of change of pressure, in the example , , , , , , , This calculation is repeated point by point until a second-order difference sequence of 998 signed values ​​is obtained. The units of these values ​​are all megapascals (MPA). Table 1 lists the pressure values, first-order difference values, and second-order difference values ​​of the first 10 sampling points collected in this embodiment. Table 1 shows the quantization process of the initial data. The second-order difference value reflects the acceleration characteristics of the pressure change rate. The advantage of this formula is that by calculating the difference twice consecutively, the trend term in the pressure signal is removed, highlighting the inflection point of transient fluctuations and providing a clear basis for positive and negative sign changes for subsequent zero-crossing detection. Each value in the calculated second-order difference sequence represents the magnitude of the pressure acceleration, for example... This indicates that the rate of pressure change changes from positive to negative and accelerates. This value is directly related to the subsequent zero-crossing point marker. That is, the position where the second-order difference sign changes from positive to negative or from negative to positive is the zero-crossing point, and finally the complete pressure difference sequence is obtained.

[0027] Table 1. Example of Pressure Acquisition and Differential Calculation , The generation submodule, based on the aforementioned pressure differential sequence, compares the signs of two adjacent differential values ​​in the sequence. For example, the third second-order differential value, -0.005 MPa, is negative, and the fourth second-order differential value, 0.000 MPa, is considered an unsigned change and temporarily disregarded. The comparison continues between the fourth 0.000 and the fifth 0.005 MPa; the fifth is positive, but it's necessary to find the position where the sign changes from positive to negative or vice versa. In this embodiment, the second second-order differential... The values ​​between 0.000 and the third -0.005 can be ignored as the second value is considered zero. The key is the third negative number -0.005 and the fourth 0.000. The fourth is non-negative and considered positive. Therefore, the change from negative to positive for the third and fourth values ​​marks a zero crossing. Similarly, for the fifth 0.005 and the sixth 0.000, the sixth is non-negative and considered positive, so there is no change. The sixth 0.000 and the seventh 0.005 remain unchanged. The seventh 0.005 and the eighth -0... The zero-crossing point is marked by the change from positive to negative when .001 appears. The zero-crossing points are marked by the change from negative to positive when the 8th -0.001 and the 9th 0.001, and the zero-crossing points are marked by the change from positive to negative when the 9th 0.001 and the 10th -0.005. Multiple zero-crossing point positions are obtained. 64 data points are extracted before and after each zero-crossing point. For example, taking the zero-crossing point between the 3rd and 4th zero-crossing points as the center, 64 sampling points before and after the corresponding time position of the zero-crossing point in the original pressure sequence are taken, for a total of 129 data points to form a waveform segment. If the zero-crossing point is located at the edge of the sequence, zeros are added. All waveform segments are combined into a dataset. In this example, 12 zero-crossing points are identified from 1000 sampling points, generating 12 waveform segments, each with 129 points. Finally, a transient waveform snapshot is obtained. This snapshot contains multiple sets of pressure transient change curves, reflecting the pressure oscillation characteristics when the water pump starts or stops or the load changes abruptly.

[0028] The integrated output submodule synchronously collects torque and speed values. The torque sensor is installed at the motor-pump coupling, and the speed is obtained through an encoder. In this embodiment, the torque measured at a certain moment is 45.6 Nm and the speed is 2980 rpm. The torque value of 45.6 and the speed value of 2980 are combined into a two-dimensional vector (45.6, 2980) as the load operating point vector. The first waveform segment data array with a length of 129 floating-point numbers is selected from the 12 waveform segments of the transient waveform snapshot. It is encapsulated in the same data package as the two-dimensional vector and stored in a structure format, containing two fields: waveform data array and vector components. This generates waveform load coupling data, which will be used for subsequent matching calculations. In this embodiment, another 11 segments and their corresponding load vectors at certain moments are also encapsulated, for a total of 12 sets of waveform load coupling data, all of which are output to the pre-locking matching module.

[0029] For example, the advance locking matching module includes: The waveform matching submodule receives each transient waveform snapshot and load operating point vector from the aforementioned 12 sets of waveform load coupling data. The standard feature code library pre-stores 10 typical operating condition reference waveforms, each with a length of 128 points on the standard time axis. For a given set of measured waveform snapshots (129 points) and reference waveforms, time axis scaling is aligned using dynamic time warping. The core cumulative distance formula is: ,in The measured waveform is the first Point Amplitude Compared with the reference waveform Point Amplitude The Euclidean distance, in megapascals (MPA), Take numbers from 1 to 129. Choose from 1 to 128, initial conditions In this embodiment, the first point of the measured waveform Reference waveform first point ,but , ;for , The first column is optional. ,but , The minimum value of the previous cell is ,in , , If we take the minimum value of 0.002, then... Calculate successively until The minimum cumulative distance value of 0.856 MPa was obtained. The advantage of this formula is that by allowing the time axis to stretch and align the waveform, it overcomes the waveform stretching or compression caused by speed fluctuations, allowing the distance value to truly reflect the similarity of the waveform shape. The calculated value... This is the waveform matching distance value, which is compared with a preset distance threshold of 1.000 MPa. If it is less than the threshold, the waveform matching is considered successful.

[0030] The phase alignment submodule, based on the waveform matching distance value of 0.856 MPa and the corresponding load operating point vector (45.6, 2980), extracts the power factor angle from the load operating point vector. The power factor, measured by a power factor meter, is 0.85 (lagging). The inverse cosine is then calculated... Based on the reference waveform number 1 corresponding to the waveform matching distance value of 0.856 MPa, the standard power factor angle of reference waveform 1 is retrieved from the standard feature code library. Formula for calculating phase offset Substituting into The advantage of this formula is that it quantifies the phase offset by the angle difference, providing a basis for locking judgment. The calculated 1.29° is compared with the preset angle threshold of 2.00°. If it is less than the threshold, the phase is consistent.

[0031] The lock determination submodule, based on the waveform matching distance value of 0.856 MPa and the phase offset of 1.29 degrees, with a preset distance threshold of 1.000 MPa and a preset angle threshold of 2.00 degrees, determines whether 0.856 is less than 1.000. If the result is yes, it then determines whether 1.29 is less than 2.00. If the result is also yes, when both comparison results indicate a less than relationship, it outputs a lock success flag, which is a Boolean value True plus the reference pump number 1, indicating that the current operating condition has successfully locked with reference waveform number 1. The lock flag is then passed to the load vector projection calculation module.

[0032] For example, the load vector projection calculation module block includes: The shaft angle calculation submodule extracts the reference pump number 1 from the successful lock identifier. Based on number 1, it retrieves the corresponding load operating point vector from the preset reference pump data table. This table records the torque and speed of the reference pump under rated operating conditions. The torque of reference pump 1 is T = 48.0 N·m, and the speed is... The formula for the reference axis angle is: Substitute The advantage of this formula is that it maps the two-dimensional load vector to a unit direction angle, providing a reference axis for subsequent orthogonal projection. The calculated result of 0.932° is used for subsequent projection calculations.

[0033] The orthogonal calculation submodule calls the reference axis angle of 0.932 degrees and the vector of the current pump's load operating point. First, calculate the unit vector of the reference axis. Projection scalar formula Projection vector Projection deviation formula (Rounded to 2980), current pump amplitude angle , the angle deviation formula The advantage of this formula lies in separating the direction and magnitude of the load vector deviation from the reference through orthogonal decomposition. The projected deviation reflects the degree of amplitude deviation, and the angular deviation reflects the degree of directional deviation. The calculation results... and Used for subsequent sorting locking.

[0034] For example, the wheel anchoring adjustment module includes: The projection deviation sorting submodule receives the projection deviation of 2980 and the angle deviation of 0.055 degrees, and starts the anchoring program. Assuming there are three pumps in the system (pump A, pump B, and pump C), the projection deviation of pump A is 2980, the projection deviation of pump B is 3120, and the projection deviation of pump C is 2760. It iterates through the values ​​of all pumps, compares 2980 of pump A with 3120 of pump B, selects the smaller 2980, and then compares it with 2760 of pump C. Since 2760 is less than 2980, pump C corresponding to the minimum value of 2760 is selected. Fine-tuning and locking are performed on pump C, and its number is recorded as C. The angle deviation data is transmitted to obtain the locked pump number C, which is subsequently used for the reference axle load calibration.

[0035] The angle deviation locking submodule, based on locking pump serial number C, recalibrates the reference axis to point towards pump C. The original reference axis angle of 0.932 degrees is adjusted to the load vector direction of pump C. The torque of pump C is 44.0 Nm, the speed is 3010 rpm, and the angle is... The new reference axis angle is set to 0.838 degrees. The angle deviation of all pumps is traversed. The angle deviation of pump A is 0.055 degrees, the angle deviation of pump B is 0.120 degrees, and the angle deviation of pump C itself is 0 degrees (coinciding with the reference axis). The angle deviations of each pump are compared pairwise. The 0.055 of pump A is compared with the 0.120 of pump B and 0.055 is selected. Then it is compared with the 0 of pump C. 0 is the smallest. The pump C with the smallest angle deviation is selected. Fine-tuning and locking are performed on pump C and its number C is recorded to obtain the reference locked pump number C.

[0036] The full-lock iterative loop submodule iteratively calls the above projection deviation sorting and angle deviation locking steps. In the first round, pump C is locked, and the set of locked pumps is A and B. The locking status is updated using the locked pump number C and the reference locked pump number C. In the second round, for the locked pumps, the projection deviation sorting is repeated from pumps A and B. The projection deviation of pump A is 2980, and that of pump B is 3120. Pump A is selected and locked. Then, the reference axis is recalibrated in the direction of pump A. The amplitude of pump A is 0.877 degrees. The angle deviation of the locked pump B is calculated and locked. This process is repeated until all three pumps are locked. Combining the received transient waveform snapshot (original 12 sets of waveforms), the target locking frequency (the pump operating frequency determined by the locking process, in this example, 49.8 Hz for pump C, 50.1 Hz for pump A, and 50.3 Hz for pump B) and the corresponding power factor angle feature combination (31.79 degrees, 32.15 degrees, and 30.98 degrees), the complete transient waveform snapshot, target locking frequency, and power factor angle feature combination data are obtained.

[0037] For example, the feature dictionary self-learning module includes: The boundary feature extraction submodule performs normalization processing on the amplitude sequence of a waveform segment in the aforementioned transient waveform snapshot. It divides the pressure value at each point by the maximum value of that segment. For example, if the maximum pressure of a segment is 0.315 MPa, dividing each point by 0.315 yields normalized values ​​between 0 and 1. The box-counting method is then used to calculate the fractal dimension, using the following formula: ,in The side length of the box (in terms of the number of sampling points). The number of boxes required to cover the waveform curve is actually calculated as follows: Each sampling point interval corresponds to the number of boxes. , , , , ,calculate They are respectively , They are respectively linear regression slope The information entropy is calculated using Shannon entropy, and the formula is: ,in The number of intervals into which the normalized magnitude is equally divided. For falling in the first The probability of each interval is calculated. In this embodiment, the normalized amplitude sequence has a total of 129 points, and the frequency of each interval is obtained. probability In order ,calculate Bit, Boundary Representation Formula The normalization is performed linearly, and the fractal dimension ranges from [specific value]. Information entropy range ,but , , The advantage of this formula lies in the fact that fractal dimension characterizes the roughness of the waveform, while information entropy characterizes its complexity. The fusion of these two elements forms the boundary representativeness, which can effectively distinguish between normal fluctuations and abnormal transients. The calculation results... Used for filtering boundary-preserving entries in subsequent clustering.

[0038] The feature fusion and clustering submodule receives the above data and, based on the second-order difference zero-crossing sequence of the transient waveform snapshot, extracts the number of zero-crossing points for each of the 12 waveform segments (15, 18, 12, 20, 16, 14, 17, 13, 19, 11, 22, and 10 in the examples), and executes the K-means clustering algorithm, setting the number of clusters. Initially, the centroids were randomly set to 15, 18, and 12. After iterative stabilization, three candidate clusters were obtained: Cluster 1 contained sample points 15, 16, 14, 17, 13, and 11; Cluster 2 contained 18, 20, 19, and 22; and Cluster 3 contained 12 and 10. Within each candidate cluster, the entries with the highest boundary representativeness were selected as boundary-retained entries. The boundary representativenesses of the samples in Cluster 1 were 0.581, 0.590, 0.575, 0.565, 0.560, and 0.543, with the maximum value of 0.590 being retained. In Cluster 2, the values ​​were 0.602, 0.598, 0.610, and 0.588, with the maximum value of 0.610 being retained. In Cluster 3, the values ​​were 0.550 and 0.530, with the maximum value of 0.550 being retained. The geometric mean of the target locking frequency was calculated for the non-boundary-retained entries within each cluster. , For the first The frequency (in Hertz) of each non-boundary-reserved entry. Let be the number of entries in the set. Cluster 1 has 5 non-reserved frequencies: 49.8, 50.1, 50.2, 49.9, and 50.0. The product of these frequencies is... ,calculate multiplied by have to multiplied by have to multiplied by have to The fifth root is obtained Hz, geometric mean of three non-reserved frequencies of cluster 2: 50.3, 50.5, and 50.4 Hz Hz, cluster 3, a non-reserved frequency of 50.0 geometric mean Hz, calculate the vector median of the power factor angle feature. After sorting the non-reserved angles of cluster 1 (31.79, 32.15, 30.98, 31.50, 31.20), the median is 31.50 degrees. The median of cluster 2 is 32.00 degrees, and the median of cluster 3 is 30.50 degrees. Perform point-by-point median calculation on the waveform amplitude of the non-boundary reserved entries to obtain the fused waveform snapshot. The fused waveform snapshot, fused frequency 50.00 Hz, fused phase angle 31.50 degrees, and boundary reserved entry index (corresponding to the second of the original 12) are passed to the fault trend monitoring module.

[0039] The fault trend monitoring submodule receives a fused waveform snapshot, a fused frequency of 50.00 Hz, a fused phase angle of 31.50 degrees, and a boundary-preserving entry index 2. It performs the Welch method on the fused waveform snapshot to calculate the power spectral density using a Hanning window with a window length of 64 points, an overlap of 32 points, and 128 FFT points, obtaining a power spectral density sequence. The power spectral centroid frequency is then extracted using the formula... ,in For the first Frequency points (unit: Hertz) This corresponds to the power spectral density value (unit: MPa² / Hz). (Number of positive frequency points) In this embodiment, the power spectral density sequence is calculated using the Welch method. Taking the first group of fused waveforms as an example, the frequency points... From 0 to 500 Hz, The calculations are listed in Table 2 (only the first 5 points are shown): , , , , ..., calculate the numerator and denominator have to The advantage of this formula is that the centroid frequency reflects the frequency band where signal energy is concentrated, and it shifts to higher frequencies as cavitation progresses. Subsequently, a centroid frequency sequence [45.2, 45.5, 45.8, 46.0, 46.1, 45.9, 46.3, 46.5, 46.7, 46.9] was calculated from 10 consecutive sets of data, using the first-order difference. The values ​​are [0.3, 0.3, 0.2, 0.1, -0.2, 0.4, 0.2, 0.2, 0.2], representing the proportion of positive values. Similarly, the slopes of the first-order difference of the fused phase angle are all greater than the preset threshold of 0.2. This result indicates that the centroid frequency shows an upward trend and the phase angle continues to increase, triggering a cavitation early warning signal. The formula obtained... The positive percentage of Hz and subsequent differential values ​​together constitute the early warning criterion. For each item that triggers an early warning, the boundary representativeness is accumulated from 0.581 to the cumulative degradation integral value. The initial integral value is 0, and 0.581 is added for each trigger. When the cumulative degradation integral value exceeds the set threshold of 5.0 (for example, the integral reaches 5.229 after 9 triggers), the corresponding boundary retained item index 2 is marked as the core degradation feature and retained in the feature dictionary. The retained core degradation feature item index 2, along with the corresponding fusion frequency of 50.00 Hz, fusion phase angle of 31.50 degrees, and fusion waveform snapshot, are transmitted to the original scheme feature dictionary self-learning module for overwriting and replacement.

[0040] The above description only illustrates the preferred embodiments of the present invention and should not be construed as limiting the scope of the claims. The present invention is not limited to the above embodiments, and variations in its specific structure are permitted. All modifications made within the scope of the independent claims of this invention are also within the scope of protection of this invention.

Claims

1. A novel water pump pressure controller, characterized in that, include: The transient waveform and vector acquisition module is used to acquire real-time pressure values ​​and perform second-order differential, detect zero-crossing points and capture data to form transient waveform snapshots, acquire torque and speed to form load operating point vectors, and transmit the two to the advance locking matching module; The pre-lock matching module is used to receive transient waveform snapshots and load operating point vectors, call standard feature codes to perform dynamic time warping comparison distance values, acquire power factor angle phase alignment, output a lock success flag, and transmit the load operating point vector and lock success flag to the load vector projection calculation module. The load vector projection calculation module is used to receive the load working point vector and the locking success mark, select the reference pump to calculate the reference axis angle, perform orthogonal projection to obtain the projection deviation and angle deviation, and transmit the two to the wheel anchoring adjustment module. The wheel anchoring adjustment module is used to receive the projection deviation and angle deviation and start anchoring, sort the projection deviation to select the minimum pump fine-tuning lock, recalibrate the reference axis, sort the angle deviation to select the minimum pump fine-tuning lock, repeat until full lock, and transmit the transient waveform snapshot, target locking frequency and power factor angle feature combination to the feature dictionary self-learning module.

2. The novel water pump pressure controller according to claim 1, characterized in that, It also includes a feature dictionary self-learning module, which is used to receive transient waveform snapshots, target locking frequency and power factor angle feature combinations and add them to the feature dictionary. When the entries are full, the earliest or low frequency record is replaced.

3. A novel water pump pressure controller according to claim 2, characterized in that, The transient waveform snapshot includes a pressure change sequence and a critical time point, and the load operating point vector includes mechanical torque and rotational speed.

4. A novel water pump pressure controller according to claim 2, characterized in that, The successful locking indicator includes a similarity metric and a phase synchronization value, and the projection deviation includes the projection distance and the deviation direction.

5. A novel water pump pressure controller according to claim 2, characterized in that, The angle deviation includes the angle difference and the angle sign; the target locking frequency includes the locking frequency value and the frequency tolerance; the power factor angle feature combination includes the phase feature and the combination weight; and the feature dictionary includes the feature record table and the replacement strategy.

6. A novel water pump pressure controller according to claim 1, characterized in that, The transient waveform and vector acquisition module includes: The pressure differential acquisition submodule acquires a real-time pressure value sequence with a pressure value sampling interval of 1 millisecond. The sequence contains 1000 sampling points. The difference between each pressure value and the pressure value at the previous moment is calculated to obtain a first-order difference sequence. Then, the difference between adjacent elements of the first-order difference sequence is calculated to obtain a second-order difference sequence. The second-order difference sequence contains 998 signed values, thus obtaining the pressure differential sequence. The segmentation generation submodule, based on the pressure differential sequence, compares the signs of two adjacent differential values ​​in the sequence. When the sign changes from positive to negative or from negative to positive, the position is marked as a zero-crossing point. 64 data points before and after each zero-crossing point are extracted to generate waveform segments. All waveform segments are combined into a dataset to obtain a transient waveform snapshot. The integrated output submodule collects torque and speed values, combines the torque and speed values ​​into a two-dimensional vector as the load operating point vector, and encapsulates the waveform data array and vector components into the same data packet based on the transient waveform snapshot and the load operating point vector to generate waveform load coupling data.

7. A novel water pump pressure controller according to claim 1, characterized in that, The prior locking and matching module includes: The waveform matching submodule receives transient waveform snapshots and load operating point vectors, aligns the transient waveform snapshots with the reference waveform in the standard feature code through dynamic time warping, calculates and accumulates the Euclidean distance point by point, takes the minimum accumulated distance value, and generates the waveform matching distance value. The phase alignment submodule extracts the power factor angle from the load operating point vector based on the waveform matching distance value and the load operating point vector. It then retrieves the corresponding standard power factor angle from the standard feature code according to the waveform matching distance value, calculates the difference between the power factor angle and the standard power factor angle, and obtains the phase offset. The lock determination submodule compares the waveform matching distance value with a preset distance threshold and the phase offset with a preset angle threshold based on the waveform matching distance value and phase offset. When both comparison results indicate a less than relationship, a lock success flag is output.

8. A novel water pump pressure controller according to claim 1, characterized in that, The load vector projection calculation module includes: The axis angle calculation submodule extracts the reference pump number from the lock success identifier, retrieves the ordinate and abscissa components of the corresponding load working point vector in the preset reference pump data table based on the number, divides the ordinate component by the abscissa component and takes the arctangent to obtain the angle value as the reference axis angle, and generates the reference axis angle. The orthogonal calculation submodule calls the reference axis angle and the load working point vector, calculates the dot product of the load working point vector and the reference axis unit vector to obtain the projection scalar, multiplies the projection scalar by the reference axis unit vector to obtain the projection vector, calculates the Euclidean norm of the difference vector between the load working point vector and the projection vector as the projection deviation, and calculates the absolute value of the difference between the argument of the load working point vector and the reference axis angle as the angle deviation, thus obtaining the projection deviation and the angle deviation.

9. A novel water pump pressure controller according to claim 1, characterized in that, The wheel anchoring adjustment module includes: The projection deviation sorting submodule receives the projection deviation and angle deviation, starts the anchoring program, iterates through the projection deviation values ​​of all pumps, compares the deviation values ​​of each pump pairwise, selects the pump with the smallest deviation value, performs fine-tuning locking on the pump and records its number, transmits the angle deviation, and obtains the locked pump number. The angle deviation locking submodule recalibrates the reference axis to point to the pump based on the locked pump serial number, iterates through the angle deviation values ​​of all pumps, compares the angle deviation values ​​of each pump pairwise, selects the pump with the smallest angle deviation value, performs fine-tuning locking on the pump and records its number, and obtains the reference locked pump number. The full-lock iterative loop submodule cyclically calls the projection deviation sorting and angle deviation locking steps. Each iteration is based on the set of unlocked pumps, and the locking status is updated using the locked pump number and the reference locked pump number. The above process is repeated until all pumps are locked. The transient waveform snapshot and the target locking frequency and power factor angle feature combination are combined to obtain the transient waveform snapshot, target locking frequency and power factor angle feature combination.

10. A novel water pump pressure controller according to claim 2, characterized in that, The feature dictionary self-learning module includes: The boundary feature extraction submodule is used to perform normalization processing on the amplitude sequence of transient waveform snapshots, call the box dimension method to calculate the fractal dimension, call the Shannon entropy to calculate the information entropy, and take the mean after normalizing the fractal dimension and information entropy as the boundary representativeness. The transient waveform snapshot, boundary representativeness, target locking frequency and power factor angle features are passed to the feature fusion clustering module. The feature fusion and clustering submodule receives transient waveform snapshots, boundary representativeness, target locking frequency, and power factor angle features. Based on the second-order difference zero-crossing sequence of the transient waveform snapshot, it performs K-means clustering algorithm to obtain candidate clusters. Within each candidate cluster, it selects the entry with the maximum value based on the boundary representativeness as the boundary-retained entry. It calculates the geometric mean of the target locking frequency for the non-boundary-retained entries in the candidate cluster to obtain the fusion frequency. It calculates the vector median of the power factor angle features to obtain the fusion phase angle. It performs point-by-point median operation on the waveform amplitude of the non-boundary-retained entries to obtain the fusion waveform snapshot. It then transmits the fusion waveform snapshot, fusion frequency, fusion phase angle, and boundary-retained entry index to the fault trend monitoring module. The fault trend monitoring submodule receives fused waveform snapshots, fused frequencies, fused phase angles, and boundary retention entry indexes. It performs Welch's method on the fused waveform snapshots to calculate the power spectral density and extracts the power spectral centroid frequency. It then performs first-order difference calculations on the power spectral centroid frequency sequence and the fused phase angle sequence, respectively. It calculates the proportion of positive values ​​in the first-order difference of the power spectral centroid frequency and compares it with a threshold of 0.

8. It determines whether the slope of the first-order difference of the fused phase angle remains positive and is greater than a preset threshold. When the proportion of positive values ​​continuously exceeds 0.8 and the phase angle slope remains positive, it triggers a cavitation warning signal. It accumulates the boundary representativeness of the triggered warning entries to the degradation cumulative integral value. When the degradation cumulative integral value exceeds a set threshold, it marks the corresponding boundary retention entry index as a degradation core feature and retains it in the feature dictionary. The retained degradation core feature entry index, along with the corresponding fused frequency, fused phase angle, and fused waveform snapshot, is transmitted to the original scheme's feature dictionary self-learning module for overwriting and replacement.

Images