Intelligent valve terminal monitoring management system of hydraulic station

Abstract

The invention discloses an intelligent valve terminal monitoring management system of a hydraulic station, which relates to the technical field of industrial automation control and comprises a data acquisition module, a feature extraction module, a time-frequency domain combined diagnosis module, a fault diagnosis decision module, an alarm module, a remote monitoring module, a data storage module and a system control module. The method has the advantages that the adaptive wavelet packet decomposition technology is adopted to perform multi-band feature extraction on the operating parameters of the hydraulic station valve terminal, and the signal local energy entropy is combined to dynamically select the optimal wavelet basis function and the decomposition layer number, so that sensitive band features corresponding to faults such as valve element abrasion and valve internal leakage can be accurately identified; according to the system, through dynamic optimization of decomposition parameters and frequency band energy analysis, the false alarm rate is remarkably reduced, meanwhile, dependence on artificial experience is avoided, the system has the self-adaptive diagnosis capacity for complex working conditions, and therefore the accuracy and reliability of fault detection are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of industrial automation control, and particularly to an intelligent valve island monitoring and management system for a hydraulic station. Background Art

[0002] In modern industrial production, the hydraulic system, as a key power transmission and control device, is widely used in many fields such as mechanical manufacturing, metallurgy, mining, aerospace, etc. As one of the core components of the hydraulic system, the valve island of the hydraulic station, the stability and reliability of its operating state directly affect the performance and production efficiency of the entire system. With the continuous improvement of industrial automation, higher requirements are put forward for the monitoring and management of the valve island of the hydraulic station. On the one hand, industrial equipment is developing towards large-scale and complex directions, and the working load and operating complexity borne by the valve island of the hydraulic station have increased significantly. For example, in a large mechanical manufacturing production line, multiple hydraulic actuators work together, and the valve island needs to accurately control the action sequence, pressure, and flow of a large number of hydraulic components to ensure the efficient and stable operation of the production line. Any failure of a valve island component may lead to the stagnation of the entire production line, causing huge economic losses;

[0003] There are certain defects in the prior art. First, the prior art mostly adopts the fixed threshold comparison method, which is difficult to accurately identify the fault-sensitive frequency band, easily confuses transient and permanent faults, and has a high false alarm rate. Second, the traditional rule-based or single time-frequency analysis method cannot automatically optimize the fault fingerprint library, has low diagnostic efficiency, and is difficult to adapt to the change of working conditions. For this reason, we propose an intelligent valve island monitoring and management system for a hydraulic station. Summary of the Invention

[0004] The purpose of the present invention is to provide an intelligent valve island monitoring and management system for a hydraulic station.

[0005] To achieve the above purpose, the present invention provides the following technical solution: An intelligent valve island monitoring and management system for a hydraulic station, the monitoring and management system includes:

[0006] A data acquisition module: used to collect parameter data during the operation of the valve island of the hydraulic station, the parameter data includes oil pressure data, oil temperature data, flow data, and spool displacement data, and convert the collected physical quantities into corresponding electrical signals respectively;

[0007] A feature extraction module: perform adaptive wavelet packet transform on the electrical signals of the data acquisition module, select the optimal wavelet basis function according to the signal local energy entropy, determine the optimal decomposition level through an iterative algorithm, calculate the energy ratio of each sub-band after decomposition, and construct a fault feature vector;

[0008] Time-frequency domain joint diagnosis module: Combining historical operating condition data, the time-domain signal is converted into a time-frequency domain signal and a time-frequency image by using an adaptive generalized S-transform. A spatio-temporal attention mechanism is introduced, and feature learning is performed on the time-frequency image through a network structure combining a multi-layer perceptron and a long short-term memory network to automatically extract fault features. Through training on a large amount of historical fault data, a fault fingerprint library is generated;

[0009] Fault diagnosis decision module: Compare and analyze the fault feature vector data with the fault fingerprint library data, calculate the similarity between the two feature vectors by using a sample distance algorithm, and judge the valve island fault, type, and degree according to the similarity threshold and rules;

[0010] Alarm module: When the fault diagnosis decision module determines that there is a fault, an alarm signal is sent through an audible and visual alarm device to remind of handling the fault;

[0011] Remote monitoring module: Through 4G / 5G network communication, remote real-time monitoring is realized, and staff can view the operation and fault information and operate at the terminal;

[0012] Data storage module: Store the original data, fault feature data, fault fingerprint library data, and diagnosis result data, and manage them with a MySQL database;

[0013] System control module: According to the fault diagnosis decision result and operation instructions, adjust the parameters of the valve island actuator through a programmable logic controller to coordinate the work of each module.

[0014] As a further solution of the present invention: In the feature extraction module, the optimal wavelet basis function is selected according to the local energy entropy of the signal, and the formula for calculating the local energy entropy of the signal is as follows:

[0015]

[0016] where, E entropy represents the local energy entropy of the signal, x i represents the value of the i-th element in the original signal sequence, x j represents the value of the j-th element in the original signal sequence, N represents the length of the signal sequence, and the wavelet basis function with the smallest local energy entropy is selected;

[0017] The iterative algorithm formula for determining the optimal decomposition level through an iterative algorithm is as follows:

[0018]

[0019] where, L optimal represents the optimal decomposition level, L represents the decomposition level variable, and ΔE k represents the energy change amount of the k-th sub-band after the L-th layer of decomposition;

[0020] The formula for calculating the energy ratio of each sub - frequency band in each layer after decomposition is as follows:

[0021]

[0022] Among them, R k represents the characteristic energy ratio of the k - th sub - frequency band, E k represents the energy of the k - th sub - frequency band, E m represents the energy of the m - th sub - frequency band, and L represents the optimal decomposition layer number.

[0023] As a further solution of the present invention: in the time - frequency domain joint diagnosis module, an adaptive generalized S - transform is used to convert the time - domain signal into a time - frequency domain signal and a time - frequency image. The formula of the adaptive generalized S - transform is as follows:

[0024]

[0025] Among them, GST x (t, f) represents the result of the adaptive generalized S - transform of the signal x(τ), x(τ) represents the pre - processed signal, τ represents the time variable, t represents the current time, f represents the frequency, σ(t) is adaptively adjusted according to the local characteristics of the signal. The specific local characteristics of the signal include the frequency change rate of the local signal, the local energy concentration degree, and the mutation degree of the local signal;

[0026] The adaptive adjustment formula of σ(t) in the generalized S - transform is as follows:

[0027]

[0028] Among them, α1, β1 and γ1 are weight coefficients, which are determined by training with historical data.

[0029] The formula for calculating the time - attention weight in the introduced spatio - temporal attention mechanism is as follows:

[0030]

[0031] Among them, a t represents the attention weight at time t, w t represents the learnable weight vector, h t represents the feature vector at time t, s represents the index variable traversing the time series, taking values from 1 to T, and T represents the total length of the time series, h s represents the feature vector at time s.

[0032] As a further solution of the present invention: the data acquisition module collects parameters through the following sensors:

[0033] Piezo - resistive pressure sensor: oil pressure data;

[0034] Thermocouple temperature sensor: oil temperature data;

[0035] Turbine flow sensor: flow rate data;

[0036] Magnetostrictive displacement sensor: spool displacement data;

[0037] The above-mentioned multiple sensors preprocess the collected analog electrical signals through the built-in signal conditioning circuit, amplification, and filtering, and then transmit them to the feature extraction module in the form of differential signals through shielded twisted pairs.

[0038] As a further solution of the present invention: The fault diagnosis and decision-making module compares and analyzes the fault feature vector data with the fault fingerprint database data, specifically as follows:

[0039] Let the fault feature vector be X = (x1, x2,..., x n ), and let a certain feature vector in the fault fingerprint database be Y = (y1, y2,..., y n ), where n represents the dimension of the feature vector. The sample distance algorithm is used to calculate the similarity between the two feature vectors. The calculation formula of the sample distance algorithm is as follows:

[0040]

[0041] Among them, D M (X, Y) represents the sample distance between the fault feature vector X and the feature vector Y in the fault fingerprint database. S represents the covariance matrix of all feature vectors in the fault fingerprint database, and S -1 represents its inverse matrix. α represents the introduced adjustment coefficient, which is used for the influence of the sum of the sample distance and the absolute difference, and its value range is (0, 1), and it is determined by experimental optimization according to the actual application scenario;

[0042] Using grid search (Grid_Search) combined with cross-validation, test different α values (0.1 to 0.9, step size 0.1) on the historical data set, and select the α value that minimizes the false alarm rate;

[0043] The relationship between the similarity S and the sample distance is as follows:

[0044]

[0045] By traversing all the feature vectors in the fault fingerprint database, calculating the similarity between the fault feature vector and each fingerprint database feature vector to obtain a set of similarity values, and comparing these similarity values with a preset similarity threshold. When there is a similarity value greater than the threshold, it is considered that the current operating state of the valve island in the hydraulic station matches the corresponding fault type in the fault fingerprint database. According to the fault type and severity information corresponding to the matching fingerprint database feature vector, the current fault situation of the valve island is judged. When all similarity values are not greater than the threshold, it is considered that the current operating state of the valve island in the hydraulic station does not match any known fault types in the fault fingerprint database. At this time, it is determined that the valve island is in a normal operating state or a new type of fault not included in the fault fingerprint database has occurred.

[0046] As a further aspect of the present invention: The fault diagnosis and decision-making module is connected to the alarm module through a hard wire. When a fault occurs, an alarm instruction is issued. When the fault diagnosis and decision-making module determines that there is a fault in the valve island of the hydraulic station, it converts the alarm signal into a level signal through the internal digital output interface and directly transmits it to the alarm module through the hard wire. After the control circuit in the alarm module receives this level signal, it triggers the audible and visual alarm device. The light flashes at a frequency of 3 times per second, and a beeping sound of 80 dB is emitted for alarm.

[0047] As a further aspect of the present invention: The remote monitoring module is connected to the data storage module through a 4G / 5G network for the terminal to obtain data for viewing and analysis. The remote monitoring module accesses the 4G / 5G network through the built-in 4G / 5G communication module. After the data storage module packs and encrypts information such as the stored original data, fault feature data, fault fingerprint database data, and diagnosis results, it sends them to the remote monitoring module according to the MQTT protocol. The remote monitoring module then decrypts and unpacks the received data and transmits it to the remote terminal through the network. The monitoring software installed on the remote terminal parses and displays the data. Staff can view the operating data and fault information of the system in real time through the operation interface and perform corresponding analysis and operations.

[0048] As a further aspect of the present invention: The system control module is connected to the actuator of the valve island in the hydraulic station through an electrical control cable to adjust the operating parameters. The system control module encodes and amplifies the control signal according to the result of the fault diagnosis and decision-making module and the operation instructions sent by the staff through the remote monitoring module / local operation interface, and transmits the control signal to the actuator of the valve island in the hydraulic station through the electrical control cable. After the control unit in the actuator receives the control signal, it adjusts its own working state according to the signal instruction.

[0049] Adopting the above technical solutions, compared with the prior art, the beneficial effects of the present invention are as follows:

[0050] 1. By adopting the adaptive wavelet packet decomposition technology, the present invention extracts multi-band features of the operating parameters of the valve island in the hydraulic station, and dynamically selects the optimal wavelet basis function and decomposition layer number in combination with the local energy entropy of the signal. It can accurately identify the sensitive band features corresponding to faults such as spool wear and internal valve leakage. By decomposing the signal into different frequency bands and calculating the energy proportion of each sub-band, it can effectively distinguish temporary faults (such as instantaneous oil pressure fluctuations) from permanent damages (such as spool jamming). For sudden changes in high-frequency band energy (such as anomalies in the 5kHz - 10kHz frequency band), it can be determined as permanent faults such as spool wear, while fluctuations in low-frequency band energy can be attributed to instantaneous interference. Compared with the traditional threshold comparison method that relies on manual experience to set fixed thresholds, this system significantly reduces the false alarm rate through dynamic optimization of decomposition parameters and frequency band energy analysis, and at the same time avoids relying on manual experience, enabling the system to have the adaptive diagnosis ability for complex working conditions, thereby improving the accuracy and reliability of fault detection;

[0051] 2. By constructing a time-frequency domain joint diagnosis model, the present invention uses the generalized S transform to convert the time-domain signal into a time-frequency image, and combines a spatio-temporal attention mechanism for feature learning. The system can automatically extract fault fingerprints based on historical working condition data and establish a dynamic knowledge base. By integrating a multi-layer perceptron (capturing local time-frequency features) and a long short-term memory network (capturing temporal correlation), it realizes deep learning and intelligent classification of fault patterns. Compared with traditional methods based on rule bases or single time-frequency analysis, this system continuously optimizes the fault fingerprint library through a self-evolution mechanism, can adapt to the fault evolution laws under different working conditions, accurately identify the fault types and their severity. In addition, through the weight assignment of the spatio-temporal attention mechanism, the system can automatically focus on key fault areas, significantly improving the diagnosis efficiency in complex scenarios, and providing a complete intelligent decision support scheme for predictive maintenance of industrial equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 It is a schematic diagram of the system flow in the embodiment of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0053] The following further describes the specific embodiments of the present invention with reference to the drawings. It should be noted here that the description of these embodiments is for helping to understand the present invention, but does not limit the present invention.

[0054] In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0055] Please refer to the attached Figure 1 , an intelligent valve island monitoring and management system for a hydraulic station of the present invention, the monitoring and management system includes:

[0056] Data acquisition module: It is used to collect parameter data during the operation of the valve island in the hydraulic station. The parameter data includes oil pressure data, oil temperature data, flow rate data, and spool displacement data, and converts the collected physical quantities into corresponding electrical signals respectively;

[0057] Feature extraction module: It performs adaptive wavelet packet transform on the electrical signals of the data acquisition module, selects the optimal wavelet basis function according to the local energy entropy of the signal, determines the optimal decomposition level through an iterative algorithm, calculates the energy ratio of each sub-band after decomposition, and constructs a fault feature vector;

[0058] Time-frequency domain joint diagnosis module: Combining historical working condition data, it uses an adaptive generalized S transform to convert the time-domain signal into a time-frequency domain signal and a time-frequency image, introduces a spatio-temporal attention mechanism, performs feature learning on the time-frequency image through a network structure combining a multi-layer perceptron and a long short-term memory network, automatically extracts fault features, and generates a fault fingerprint library through the training of a large amount of historical fault data;

[0059] Fault diagnosis decision module: It compares and analyzes the fault feature vector data with the fault fingerprint library data, calculates the similarity between the two feature vectors using the sample distance algorithm, and judges the valve island fault, type, and degree according to the similarity threshold and rules;

[0060] Alarm module: When the fault diagnosis decision module determines that there is a fault, it sends an alarm signal through an audible and visual alarm device to remind to handle the fault;

[0061] Remote monitoring module: Through 4G / 5G network communication, it realizes remote real-time monitoring, and staff can view the operation and fault information and operate at the terminal;

[0062] Data storage module: It stores the original data, fault feature data, fault fingerprint library data, and diagnosis result data, and manages them with a MySQL database;

[0063] System control module: According to the fault diagnosis decision result and operation instructions, it adjusts the parameters of the valve island actuator through a programmable logic controller and coordinates the work of each module.

[0064] In an embodiment of the present invention: In the feature extraction module, the optimal wavelet basis function is selected according to the local energy entropy of the signal, and the formula for calculating the local energy entropy of the signal is as follows:

[0065]

[0066] Among them, E entropy represents the local energy entropy of the signal, x i represents the value of the i-th element in the original signal sequence, x j represents the value of the j-th element in the original signal sequence, N represents the length of the signal sequence, and the wavelet basis function with the smallest local energy entropy is selected;

[0067] The iterative algorithm formula for determining the optimal decomposition level through an iterative algorithm is as follows:

[0068]

[0069] Among them, L optimal represents the optimal decomposition level, L represents the decomposition level variable, and ΔE k represents the energy change of the k-th sub-band after the L-th layer of decomposition;

[0070] The formula for calculating the energy ratio of each sub-band in each layer after decomposition is as follows:

[0071]

[0072] Among them, R k represents the characteristic energy ratio of the k-th sub-band, E k represents the energy of the k-th sub-band, E m represents the energy of the m-th sub-band, and L represents the optimal decomposition level.

[0073] In an embodiment of the present invention: in the time-frequency domain joint diagnosis module, an adaptive generalized S transform is used to convert the time-domain signal into a time-frequency domain signal and a time-frequency image. The formula for the adaptive generalized S transform is as follows:

[0074]

[0075] Among them, GST x (t, f) represents the result of the adaptive generalized S transform of the signal x(τ), x(τ) represents the preprocessed signal, τ represents the time variable, t represents the current time, f represents the frequency, σ(t) is adaptively adjusted according to the local characteristics of the signal, and the local characteristics of the signal specifically include the frequency change rate of the local signal, the local energy concentration degree, and the mutation degree of the local signal;

[0076] The adaptive adjustment formula of σ(t) in the generalized S transform is as follows:

[0077]

[0078] Among them, α1, β1, and γ1 are weight coefficients, which are determined by training with historical data.

[0079] The formula for calculating the time attention weight in the introduced spatio-temporal attention mechanism is as follows:

[0080]

[0081] Among them, a t represents the attention weight at time t, w t represents the learnable weight vector, ht denotes the feature vector at time t, s denotes the index variable traversing the time series, taking values from 1 to T, where T represents the total length of the time series, and h s denotes the feature vector at time s.

[0082] In an embodiment of the present invention: The data acquisition module acquires parameters through the following sensors:

[0083] Piezoelectric pressure sensor: oil pressure data;

[0084] Thermocouple temperature sensor: oil temperature data;

[0085] Turbine flow sensor: flow rate data;

[0086] Magnetostrictive displacement sensor: spool displacement data;

[0087] The above-mentioned multiple sensors transmit the collected analog electrical signals to the feature extraction module in the form of differential signals through shielded twisted pairs after preprocessing by the built-in signal conditioning circuit, amplification, and filtering.

[0088] In an embodiment of the present invention: The fault diagnosis decision module compares and analyzes the fault feature vector data with the fault fingerprint database data, specifically as follows:

[0089] Let the fault feature vector be X = (x1, x2, …, x n ), and let a certain feature vector in the fault fingerprint database be Y = (y1, y2, …, y n ), where n represents the dimension of the feature vector. The similarity between the two feature vectors is calculated using the sample distance algorithm, and the calculation formula of the sample distance algorithm is as follows:

[0090]

[0091] Among them, D M (X, Y) represents the sample distance between the fault feature vector X and the feature vector Y in the fault fingerprint database, S represents the covariance matrix of all feature vectors in the fault fingerprint database, and S -1 represents its inverse matrix, α represents the introduced adjustment coefficient, which is used for the influence of the sum of the sample distance and the absolute difference, and its value range is (0, 1), and it is determined by experimental optimization according to the actual application scenario;

[0092] Using grid search (GridSearch) combined with cross-validation, different α values (from 0.1 to 0.9, step size 0.1) are tested on the historical dataset, and the α value that minimizes the false alarm rate is selected;

[0093] The relationship between the similarity S and the sample distance is as follows:

[0094]

[0095] By traversing all the feature vectors in the fault fingerprint database, calculating the similarity between the fault feature vector and each fingerprint database feature vector to obtain a set of similarity values, comparing these similarity values with a preset similarity threshold. When there is a similarity value greater than the threshold, it is considered that the operating state of the current valve island of the hydraulic station matches the corresponding fault type in the fault fingerprint database. According to the fault type and severity information corresponding to the matched fingerprint database feature vector, the current fault situation of the valve island is judged. When all similarity values are not greater than the threshold, it is considered that the operating state of the current valve island of the hydraulic station does not match the known fault types in the fault fingerprint database. At this time, it is determined that the valve island is in a normal operating state or a new type of fault not included in the fault fingerprint database has occurred.

[0096] In an embodiment of the present invention: The fault diagnosis decision module is hard-wired to the alarm module and issues an alarm instruction during a fault. When the fault diagnosis decision module determines that there is a fault in the valve island of the hydraulic station, it converts the alarm signal into a level signal through the internal digital output interface and directly transmits it to the alarm module through the hard wire. After the control circuit in the alarm module receives this level signal, it triggers the audible and visual alarm device, and the light flashes at a frequency of 3 times / s and emits a beeping sound of 80 dB for alarm.

[0097] In an embodiment of the present invention: The remote monitoring module is connected to the data storage module through a 4G / 5G network for the terminal to obtain data for viewing and analysis. The remote monitoring module accesses the 4G / 5G network through the built-in 4G / 5G communication module. After the data storage module packs and encrypts information such as the stored original data, fault feature data, fault fingerprint database data, and diagnosis results, it sends them to the remote monitoring module according to the MQTT protocol. The remote monitoring module then decrypts and unpacks the received data and transmits it to the remote terminal through the network. The monitoring software installed on the remote terminal parses and displays the data, and the staff can view the operation data and fault information of the system in real time through the operation interface and perform corresponding analysis and operations.

[0098] In an embodiment of the present invention: The system control module is connected to the actuator of the valve island of the hydraulic station through an electrical control cable to adjust the operating parameters. The system control module encodes and amplifies the control signal according to the result of the fault diagnosis decision module and the operation instructions sent by the staff through the remote monitoring module / local operation interface, and transmits the control signal to the actuator of the valve island of the hydraulic station through the electrical control cable. After the control unit in the actuator receives the control signal, it adjusts its own working state according to the instruction of the signal.

[0099] Example 1. Please refer to the appendix Figure 1: System parameter settings, acquisition module parameters: When a valve island in a hydraulic station is operating normally, the measurement range of the piezoresistive pressure sensor is set to 0 - 30 MPa, with an accuracy of ±0.1 MPa. The measurement range of the thermocouple temperature sensor is 0 - 100 °C, with an accuracy of ±1 °C. The measurement range of the turbine flow sensor is 0 - 50 L / min, with an accuracy of ±0.5 L / min. The measurement range of the magnetostrictive displacement sensor is 0 - 100 mm, with an accuracy of ±0.1 mm;

[0100] Feature extraction module parameters: The initial search range for the number of wavelet packet decomposition layers is set to 2 - 6 layers, and the signal sequence length N is set to 2048 points;

[0101] Time-frequency domain joint diagnosis module parameters: In the generalized S transform, the frequency resolution is set to 0.05 Hz, and the time resolution is set to 0.005 s. In the spatio-temporal attention mechanism, the initial value of the learnable weight vector is randomly initialized by a uniform distribution within the interval [-1, 1];

[0102] Fault diagnosis decision module parameters: The similarity threshold is set to 0.75, and the initial value of the adjustment coefficient α in the sample distance algorithm is set to 0.4;

[0103] Simulate the occurrence of a fault, use a simulation test bench to simulate the spool wear fault. During the simulation, as the spool wear intensifies, fault characteristic signals gradually appear in the frequency band of 5 kHz - 10 kHz. The data acquisition module collects relevant parameters in real time. At a certain moment, the piezoresistive pressure sensor collects an oil pressure of 12 MPa with slight fluctuations, the thermocouple temperature sensor collects an oil temperature of 42 °C with a slight increase, the turbine flow sensor collects a flow rate of 23 L / min which is basically stable, and the magnetostrictive displacement sensor collects abnormal changes in the spool displacement. After these physical quantities are converted into electrical signals, they are preprocessed such as amplified and filtered through the signal conditioning circuit built in the sensor, and then transmitted to the feature extraction module in the form of differential signals through shielded twisted pair;

[0104] Feature extraction: After receiving the electrical signal, the feature extraction module starts to calculate the local energy entropy of the signal. Taking a signal with a duration of 1 second and a sampling frequency of 10 kHz as an example, this signal sequence contains 10000 data points. Select 2048 consecutive points from it as the sample for calculating the local energy entropy (i.e., N = 2048). According to the formula:

[0105]

[0106] Calculate for different wavelet basis functions (common wavelet basis functions such as db2, db3, sym4, etc.). After calculation and comparison, it is found that when using the db3 wavelet basis function, the local energy entropy of the signal is the smallest. Therefore, db3 is selected as the optimal wavelet basis function;

[0107] After determining the wavelet basis function, the optimal decomposition level is determined through an iterative algorithm. Starting from the initially set 2 levels, according to the formula:

[0108]

[0109] Calculate the sum of the absolute values of the energy change amounts of each sub-band at different decomposition levels;

[0110] When the decomposition level L = 2, the calculated result is

[0111] When L = 3, the calculated result is

[0112] When L = 4, the calculated result is

[0113] When L = 5, the calculated result is

[0114] When L = 6, the calculated result is

[0115] By comparison, it can be seen that when L = 4, this value is the smallest, so the optimal decomposition level is determined to be 4 levels;

[0116] Next, calculate the energy ratio of each sub-band at each layer. For the 16 sub-bands (2 4 = 16) after 4-layer decomposition, according to the formula:

[0117]

[0118] Calculate the energy ratio of each sub-band. The energy E3 of the 3rd sub-band = 0.2;

[0119] The sum of the energies of all sub-bands Then the energy ratio of this sub-band By calculating the energy ratios of all sub-bands, a fault feature vector is constructed.

[0120] For diagnostic analysis, the time-frequency domain joint diagnostic module combines historical operating condition data and performs a generalized S-transform on the collected time-domain signal. At the current time t = 0.5 s and frequency f = 8 kHz, perform a generalized S-transform on the preprocessed signal x(τ) (τ is the time variable) according to the formula:

[0121]

[0122] where σ(t) is adaptively adjusted according to the local characteristics of the signal. For the current signal, according to the frequency change rate, local energy concentration degree, and mutation degree of the local signal, determine that σ(τ) = 0.01, and the time-frequency domain signal and time-frequency image are obtained through calculation.

[0123] Introduce a spatio-temporal attention mechanism to calculate the temporal attention weights. Assume the total length of the time series is \(T = 100\), the current time is \(t = 30\), and the learnable weight vector is \(w\). t and the feature vector \(h\) at time \(t\). t After performing a dot product operation, we get \(w\). t ·\(h\). t = 0.5. For each moment \(s\) in the time series, calculate \(\exp(w\). t , \(h\). t ). Then, according to the formula:

[0124]

[0125] calculate the temporal attention weight \(a\). t Similarly, calculate the spatial attention weight (the specific calculation process is similar to that of the temporal attention weight, according to the corresponding formula). Through a network structure combining a multi-layer perceptron and a long short-term memory network, perform feature learning on the time-frequency image, automatically extract fault features, and compare them with the fault fingerprint database.

[0126] The fault judgment and fault diagnosis decision-making module compares and analyzes the fault feature vector obtained by the feature extraction module with the data in the fault fingerprint database. Assume the fault feature vector is \(X=(x_1,x_2,\cdots,x\). n ), and a certain feature vector \(Y=(y_1,y_2,\cdots,y\). n ) corresponding to the spool wear fault in the fault fingerprint database. The dimension of the feature vector is \(n = 10\). The covariance matrix \(S\) of all feature vectors in the fault fingerprint database is calculated from historical data, and its inverse matrix \(S\). -1 has also been determined. According to the sample distance algorithm formula:

[0127]

[0128] calculate the sample distance \(D\). M (X,Y)=0.3. Then, according to the relationship between similarity and sample distance:

[0129]

[0130] calculate the similarity. Since 0.769 is greater than the preset similarity threshold of 0.75, it is determined that there is a spool wear fault in the valve island.

[0131] Alarm and control response, alarm trigger. After the fault diagnosis decision-making module determines a fault, it sends an alarm instruction to the alarm module through a hard wire. The control circuit in the alarm module receives the level signal and triggers the audible and visual alarm device. The light flashes 3 times per second, and a beeping sound of 80 decibels is emitted to remind the staff.

[0132] Control adjustment: Based on the results of fault diagnosis decision-making, the system control module combines the preset control strategy and adjusts the parameters of the valve island actuator through the programmable logic controller. For example, it reduces the working pressure of the hydraulic station to 10 MPa to reduce further wear of the spool. The control signal is transmitted to the actuator of the valve island of the hydraulic station, such as electromagnetic directional valves, proportional relief valves, etc., through the electrical control cable, so that they adjust their own working states according to the instructions.

[0133] Example 2: Please refer to the appendix Figure 1 For system parameter setting (parameters related to threshold adjustment except for basic parameters), the initial similarity threshold is set to 0.8, and the adjustment coefficient α in the sample distance algorithm is adjusted to 0.5. At the same time, the trigger condition for threshold adjustment is set: when among 5 consecutive diagnostic results, 3 times are judged as suspected faults (similarity is close but does not reach the threshold), the threshold self-adaptive adjustment mechanism is started, and the threshold adjustment step size is set to 0.05, that is, the amplitude of adjusting the similarity threshold each time is 0.05.

[0134] Simulation of fault occurrence and preliminary diagnosis: Use the simulation experiment equipment to simulate the internal leakage fault of the valve. The data acquisition module collects data such as the oil pressure gradually decreasing from 12 MPa to 10 MPa, the oil temperature changing slightly and remaining at about 40 °C, and the flow rate abnormally increasing from 23 L / min to 28 L / min. These data are transmitted to the feature extraction module and the time-frequency domain joint diagnosis module after being processed. Finally, the fault diagnosis decision-making module conducts a preliminary diagnosis. In the initial several diagnoses, due to the insufficiently obvious fault characteristics, it may occur that among 5 consecutive diagnoses, 3 times are judged as suspected faults. The similarity of the first diagnosis is 0.78, the second is 0.76, the third is 0.79, the fourth is 0.81 (reaching the threshold and judged as a fault), and the fifth is 0.77. At this time, the trigger condition for threshold adjustment is met, and the self-adaptive adjustment mechanism is started.

[0135] Threshold self-adaptive adjustment process: After the fault diagnosis decision-making module detects that the trigger condition is met, it adjusts the similarity threshold according to the preset adjustment step size. Since the previous multiple diagnostic results are close to the threshold, it indicates that the current threshold may be set too high, which is not conducive to diagnosing faults timely and accurately. Therefore, the similarity threshold is reduced by 0.05 to 0.75. At the same time, the adjustment coefficient α in the sample distance algorithm is updated, and the value of α is recalculated according to the historical data and the change of the current fault characteristics through a preset algorithm (factors such as the difference between the similarity and the threshold, the change trend of the feature vector, etc.). After calculation, α is adjusted to 0.55 to more accurately measure the similarity between the fault feature vector and the data in the fault fingerprint database.

[0136] Perform fault diagnosis again with the adjusted diagnostic results, using the adjusted thresholds and parameters. Recalculate the similarity between the fault feature vector and the feature vector related to internal valve leakage in the fault fingerprint library. Suppose the calculated similarity after adjustment is 0.78. At this time, since the similarity threshold has been adjusted to 0.75 and this similarity is greater than the threshold, it is clearly determined that there is an internal valve leakage fault in the valve island. Subsequently, the system will continuously monitor the fault situation and diagnostic results. If the diagnostic results are stable within a certain period of time and conform to the actual fault situation, the adjusted thresholds and parameters will be maintained. If abnormal conditions occur again, the threshold adaptive adjustment mechanism may be triggered again to ensure the accuracy and timeliness of fault diagnosis.

[0137] According to the content of the above embodiments, by applying the adaptive wavelet packet decomposition technology to extract multi-band features of the operating parameters of the valve island in the hydraulic station, dynamically selecting the optimal wavelet basis function and decomposition level in combination with the local energy entropy of the signal, and constructing a time-frequency domain joint diagnosis model, the system can accurately identify the characteristics of the fault-sensitive frequency band, effectively distinguish between temporary faults and permanent damages, automatically extract fault fingerprints and establish a dynamic knowledge base. In actual operation, each module collaborates closely. The data acquisition module provides comprehensive operating parameters, the feature extraction module accurately analyzes the signal characteristics, the time-frequency domain joint diagnosis module deeply mines the fault patterns, the fault diagnosis decision module accurately judges the faults, the alarm module gives timely warnings, the remote monitoring module realizes convenient management, the data storage module provides data support for system optimization, and the system control module ensures the stable operation of the equipment. Through the coordinated cooperation of these technical means, the system significantly reduces the false alarm rate, avoids relying on manual experience, has the adaptive diagnosis ability for complex working conditions, can accurately identify the fault type and its severity, automatically focus on the key fault areas, and improves the accuracy, reliability and diagnostic efficiency of fault detection, providing a strong guarantee for the stable operation of the valve island in the hydraulic station and the predictive maintenance of industrial equipment, demonstrating significant technical advantages and broad application prospects.

[0138] Although the present invention is disclosed above in a preferred embodiment, it is not intended to limit the present invention. Any person skilled in the art can make possible changes and modifications without departing from the spirit and scope of the present invention. Therefore, any modification, equivalent change and decoration made to the above embodiments based on the technical essence of the present invention without departing from the technical solution of the present invention shall fall within the protection scope defined by the claims of the present invention.

Claims

1. An intelligent valve island monitoring and management system for a hydraulic station, characterized in that, The monitoring and management system includes: Data acquisition module: used to collect parameter data during the operation of the valve island in the hydraulic station. The parameter data includes oil pressure data, oil temperature data, flow rate data, and spool displacement data, and converts the collected physical quantities into corresponding electrical signals respectively; Feature extraction module: performs adaptive wavelet packet transform on the electrical signals of the data acquisition module, selects the optimal wavelet basis function according to the local energy entropy of the signal, determines the optimal decomposition level through an iterative algorithm, calculates the energy ratio of each sub-band in each layer after decomposition, and constructs a fault feature vector; Time-frequency domain joint diagnosis module: combines historical working condition data, uses an adaptive generalized S transform to convert the time-domain signal into a time-frequency domain signal and a time-frequency image, introduces a spatio-temporal attention mechanism, performs feature learning on the time-frequency image through a network structure combining a multi-layer perceptron and a long short-term memory network, automatically extracts fault features, and generates a fault fingerprint library through training on a large amount of historical fault data; Fault diagnosis decision module: compares and analyzes the fault feature vector data with the fault fingerprint library data, calculates the similarity between the two feature vectors using the sample distance algorithm, and judges the valve island fault, type, and degree according to the similarity threshold and rules; Alarm module: when the fault diagnosis decision module determines that there is a fault, it sends an alarm signal through an audible and visual alarm device to remind of handling the fault; Remote monitoring module: realizes remote real-time monitoring through 4G / 5G network communication, and staff can view the operation and fault information and operate at the terminal; Data storage module: stores the original data, fault feature data, fault fingerprint library data, and diagnosis result data, and manages them with a MySQL database; System control module: according to the fault diagnosis decision result and operation instructions, adjusts the parameters of the valve island actuator through a programmable logic controller and coordinates the work of each module.

2. The intelligent valve island monitoring and management system for a hydraulic station according to claim 1, characterized in that: In the feature extraction module, the optimal wavelet basis function is selected according to the local energy entropy of the signal, and the formula for calculating the local energy entropy of the signal is as follows: Among them, E entropy represents the local energy entropy of the signal, and x i represents the value of the i-th element in the original signal sequence, and x j represents the value of the j-th element in the original signal sequence. N represents the length of the signal sequence. Select the wavelet basis function with the minimum local energy entropy; The iterative algorithm formula for determining the optimal decomposition level through an iterative algorithm is as follows: Among them, L optimal represents the optimal decomposition level, L represents the decomposition level variable, and ΔE k represents the energy change amount of the k-th sub-band after the L-th layer of decomposition; The formula for calculating the energy ratio of each sub-band in each layer after decomposition is as follows: Among them, R k represents the characteristic energy ratio of the k-th sub-band, E k represents the energy of the k-th sub-band, E m represents the energy of the m-th sub-band, and L represents the optimal decomposition level.

3. The intelligent valve island monitoring and management system for a hydraulic station according to claim 1, characterized in that: In the time-frequency domain joint diagnosis module, an adaptive generalized S transform is used to convert the time-domain signal into a time-frequency domain signal and a time-frequency image, and the formula for the adaptive generalized S transform is as follows: Among them, GST x (t,f) represents the adaptive generalized S-transform result of the signal x(τ), x(τ) represents the preprocessed signal, τ represents the time variable, t represents the current time, f represents the frequency, and σ(t) is adaptively adjusted according to the local characteristics of the signal. The local characteristics of the signal specifically include the frequency change rate of the local signal, the local energy concentration degree, and the mutation degree of the local signal; The adaptive adjustment formula for σ(t) in the generalized S transform is as follows: Among them, α1, β1, and γ1 are weight coefficients, which are determined through training on historical data. The formula for calculating the time attention weight in the introduced spatio-temporal attention mechanism is as follows: where a t represents the attention weight at time t, w t represents the learnable weight vector, h t represents the feature vector at time t, s represents the index variable traversing the time series, taking values from 1 to T, and T represents the total length of the time series, h s represents the feature vector at time s.

4. The intelligent valve island monitoring and management system for a hydraulic station according to claim 1, characterized in that: The data acquisition module collects parameters through the following sensors: Piezo-resistive pressure sensor: oil pressure data; Thermocouple temperature sensor: oil temperature data; Turbine flow sensor: flow rate data; Magnetostrictive displacement sensor: spool displacement data; The above-mentioned multiple sensors transmit the collected analog electrical signals to the feature extraction module in the form of differential signals through a shielded twisted pair after preprocessing by an internal signal conditioning circuit, amplification, and filtering.

5. The intelligent valve island monitoring and management system of a hydraulic station according to claim 1, characterized in that: The fault diagnosis decision module compares and analyzes the fault feature vector data with the fault fingerprint library data, specifically as follows: Let the fault feature vector be X = (x1, x2, …, x n ), and let a certain feature vector in the fault fingerprint database be Y = (y1, y2, …, y n ), where n represents the dimension of the feature vector. The similarity between the two feature vectors is calculated using the sample distance algorithm, and the calculation formula of the sample distance algorithm is as follows: Among them, D M (X, Y) represents the sample distance between the fault feature vector X and the feature vector Y in the fault fingerprint database. S represents the covariance matrix of all feature vectors in the fault fingerprint database, S -1 represents its inverse matrix. α represents the introduced adjustment coefficient, which is used for the influence of the sum of the sample distance and the absolute difference, and its value range is (0, 1), and it is determined by experimental optimization according to the actual application scenario; Using grid search (Grid_Search) combined with cross-validation, different α values (from 0.1 to 0.9, with a step size of 0.1) are tested on the historical dataset, and the α value that minimizes the false alarm rate is selected; The relationship between the similarity S and the sample distance is as follows: By traversing all the feature vectors in the fault fingerprint library, calculating the similarity between the fault feature vector and each fingerprint library feature vector, a set of similarity values is obtained. These similarity values are compared with a preset similarity threshold. When there is a similarity value greater than the threshold, it is considered that the current operating state of the valve island in the hydraulic station matches the corresponding fault type in the fault fingerprint library. According to the fault type and severity information corresponding to the matched fingerprint library feature vector, the current fault situation of the valve island is judged. When all similarity values are not greater than the threshold, it is considered that the current operating state of the valve island in the hydraulic station does not match the known fault types in the fault fingerprint library. At this time, it is determined that the valve island is in a normal operating state or a new type of fault not included in the fault fingerprint library has occurred.

6. The intelligent valve island monitoring and management system of a hydraulic station according to claim 1, characterized in that: The fault diagnosis decision module is hard-wired to the alarm module and issues an alarm instruction during a fault. When the fault diagnosis decision module determines that there is a fault in the valve island of the hydraulic station, it converts the alarm signal into a level signal through the internal digital output interface and directly transmits it to the alarm module through a hard wire. After receiving this level signal, the control circuit in the alarm module triggers the audible and visual alarm device, and the light flashes at a frequency of 3 times per second, emitting a beeping sound of 80 dB for alarm.

7. The intelligent valve island monitoring and management system of a hydraulic station according to claim 1, characterized in that: The remote monitoring module is connected to the data storage module through a 4G / 5G network. The terminal obtains data for viewing and analysis. The remote monitoring module accesses the 4G / 5G network through the built-in 4G / 5G communication module. After the data storage module packs and encrypts information such as the stored original data, fault feature data, fault fingerprint library data, and diagnosis results, it sends them to the remote monitoring module according to the MQTT protocol. The remote monitoring module then decrypts and unpacks the received data and transmits it to the remote terminal through the network. The monitoring software installed on the remote terminal parses and displays the data, and the staff can view the operating data and fault information of the system in real time through the operation interface and perform corresponding analysis and operations.

8. The intelligent valve island monitoring and management system for a hydraulic station according to claim 1, wherein: The system control module is connected to the actuator of the valve island in the hydraulic station through an electrical control cable to adjust the operating parameters. The system control module encodes and amplifies the control signal according to the result of the fault diagnosis decision module and the operation instructions sent by the staff through the remote monitoring module / local operation interface, and transmits the control signal to the actuator of the valve island in the hydraulic station through the electrical control cable. After receiving the control signal, the control unit in the actuator adjusts its own working state according to the signal instruction.

Images