Pilot operated relief valve failure analysis method and system

Abstract

This invention discloses a method and system for fault analysis of pilot-operated relief valves, relating to the field of hydraulic valve fault diagnosis technology. It constructs a multi-condition dynamic response benchmark feature library, acquiring time-domain, frequency-domain, and time-frequency-domain features of the relief valve inlet pressure, pilot valve front chamber pressure, and main valve core displacement signals under healthy conditions and different degrees of blockage, spring fatigue, and valve core jamming faults through simulation and actual measurement. This forms a feature vector library with fault labels. This invention constructs a benchmark feature library for multiple operating conditions and fault types, integrating time-domain, frequency-domain, and time-frequency-domain multi-dimensional features, and combining wavelet packet energy spectrum features to accurately distinguish different fault types. This solves the problems of low fault identification and difficulty in distinguishing similar faults in traditional technologies, significantly improving the identification accuracy of latent and early-stage faults.

Description

Technical Field

[0001] This invention relates to the field of hydraulic valve fault diagnosis technology, and in particular to a fault analysis method and system for pilot-operated relief valves. Background Technology

[0002] Pilot-operated relief valves are core pressure regulating components in hydraulic systems, playing a crucial role in system pressure stabilization, overload protection, and pressure relief. Their operational stability directly determines the reliability and safety of the entire hydraulic equipment. During long-term industrial operation, relief valves are prone to typical faults such as damping orifice blockage, spring fatigue, and valve core jamming due to various factors including hydraulic oil contamination, medium aging, mechanical fatigue, and valve core wear.

[0003] Currently, the maintenance of pilot-operated relief valves in the industrial sector mainly relies on periodic inspections and reactive repairs. Condition monitoring methods are relatively limited, generally depending on manual inspections and monitoring of basic pressure parameters for fault diagnosis, which has significant technical shortcomings. Conventional monitoring can only collect basic pressure data and cannot capture subtle changes in the valve core's dynamic response, making it difficult to distinguish the characteristic differences between different fault types. It also has low accuracy in identifying early-stage faults such as minor blockages and latent spring fatigue, making it prone to missed or false diagnoses.

[0004] Meanwhile, existing technologies lack proactive diagnostic incentive mechanisms, relying solely on natural equipment operation signals for analysis. This results in unclear dynamic response characteristics of valve cores, insufficient fault feature identification, and an inability to accurately quantify the degree of damping orifice blockage and spring fatigue. In terms of fault handling, traditional systems lack tiered control logic, mostly only capable of triggering fault alarms or direct shutdowns. They lack transitional control measures for minor fault self-healing and moderate fault component switching, leading to either unnecessary equipment shutdowns impacting production efficiency or untimely fault handling causing safety hazards such as system pressure instability, pipeline shocks, and equipment overload damage.

[0005] In addition, traditional fault diagnosis models are fixed and cannot adapt to actual changes in operating conditions such as performance degradation and fluctuations in operating conditions during long-term operation of equipment. The feature database does not have the ability to iterate and update autonomously. After long-term operation, the accuracy of fault identification continues to decline, making it difficult to adapt to complex and ever-changing industrial hydraulic operating scenarios and unable to meet the dual requirements of equipment operation safety and production continuity. Summary of the Invention

[0006] To address the aforementioned technical problems, this invention provides a method and system for fault analysis of pilot-operated relief valves. The following technical solution is adopted:

[0007] The fault analysis method for pilot-operated relief valves includes the following steps:

[0008] Step 1: Construct a multi-condition dynamic response benchmark feature library. Through simulation and actual measurement, obtain the time domain, frequency domain, and time-frequency domain features of the overflow valve inlet pressure, pilot valve front chamber pressure, and main valve core displacement signal under health conditions and different degrees of blockage, spring fatigue, and valve core jamming faults, and form a feature vector library with fault labels.

[0009] Step 2: Real-time acquisition of the overflow valve inlet pressure signal, the pilot valve front chamber pressure signal, and the main valve core displacement signal; and periodically or event-triggered superimposed an active pulse diagnostic excitation signal onto the pilot valve control signal to stimulate the valve core dynamic response.

[0010] Step 3: Extract the feature vector of the real-time response signal, use the improved fuzzy C-means clustering algorithm to calculate the Mahalanobis distance between it and each cluster center in the benchmark feature library, and combine it with the damping orifice blockage sensitivity coefficient to perform fault mode identification, and output the fault type and blockage level.

[0011] Step 4: Implement graded safety interlock control according to the fault level, where L1 is high-frequency flutter self-cleaning, L2 is automatic switching to backup damping components, and L3 is safe shutdown.

[0012] Step 5: Use an incremental support vector machine to perform self-learning updates on the benchmark feature library.

[0013] Optionally, in step 1, the wavelet packet energy spectrum is decomposed into three layers using the Daubechies4 wavelet basis on the acquired pilot valve pre-cavity pressure signal. The energy proportion of the eight frequency band nodes in the third layer is extracted as the feature vector. The reconstructed signal energy of each frequency band node is calculated, the total energy is calculated, and the energy proportion of each node is extracted. The energy proportions of three nodes corresponding to the frequency ranges of 100Hz-200Hz, 200Hz-400Hz, and 400Hz-800Hz are selected as the feature vectors. This is used to distinguish between damping orifice blockage and spring fatigue.

[0014] Optionally, the generation and adaptive adjustment of the active pulse diagnostic excitation signal in step 2 includes the following sub-steps:

[0015] Step 21: Set the basic excitation waveform to a bidirectional rectangular pulse sequence, with each pulse width of 0.05 seconds, a total of four pulses, a pulse interval of 0.025 seconds, and a total excitation duration of no more than 0.3 seconds;

[0016] Step 22: Obtain the steady-state value of the main valve core displacement in real time and compare it with the reference displacement corresponding to the rated flow rate;

[0017] Step 23: Adaptively adjust the excitation amplitude based on the ratio of steady-state displacement to reference displacement: when the ratio is greater than or equal to the set maximum flow rate, the excitation amplitude is set to 5% of the rated maximum control signal value; when the ratio is less than or equal to the set minimum flow rate, the excitation amplitude is set to 20%; otherwise, it is set to 8%.

[0018] Step 24: The generated excitation signal is superimposed on the conventional control signal of the pilot valve, and the superposition method is bidirectional alternating positive and negative pulses;

[0019] Step 25: Within 0.1 seconds after the excitation ends, monitor the maximum deviation of the inlet pressure. If the maximum deviation exceeds 10% of the set pressure, mark the current excitation as invalid and automatically reduce the amplitude of the next excitation by 20%.

[0020] Optionally, step 3, which involves improving the fuzzy C-means clustering algorithm and calculating the damping orifice clogging sensitivity coefficient, includes the following sub-steps:

[0021] Step 31: Bandpass filtering and normalization are performed on the inlet pressure, pilot valve front chamber pressure and main valve core displacement signals in the real-time acquisition window to make the dimensions of each characteristic consistent.

[0022] Step 32: Extract the time-domain, frequency-domain, and time-frequency-domain features to form a real-time feature vector;

[0023] Step 33: Calculate the Mahalanobis distance between the real-time feature vector and the fault cluster center of each class in the benchmark feature library. The Mahalanobis distance is calculated by weighting the covariance matrix of the corresponding class samples.

[0024] Step 34: Calculate the fuzzy membership degree of the real-time feature vector to each fault type based on Mahalanobis distance. The membership degree function adopts the inverse square weighted form.

[0025] Step 35: Construct an improved objective function. The objective function adds a penalty term to the original fuzzy C-means clustering objective. The penalty term is the sum of the inverses of the intra-class variances of each fault sample in the two feature dimensions of damping orifice blockage sensitivity coefficient and pilot pressure fluctuation peak-to-peak value, multiplied by the penalty coefficient. The cluster centers are iteratively updated using the gradient descent method until convergence.

[0026] Step 36, calculate the damping orifice clogging sensitivity coefficient. ;

[0027] ;

[0028] in This represents the time delay corresponding to the maximum peak value of the cross-correlation function between the pilot valve inlet pressure and the main valve spool displacement. The rise time of the step response of the main valve core displacement. The standard deviation of the leading pressure, This represents the average value of the pilot pressure.

[0029] Step 37: Determine the fault level based on the maximum membership degree and the blockage sensitivity coefficient: If the maximum membership degree is greater than or equal to 0.6 and the blockage sensitivity coefficient is less than or equal to 1.2, output the fault level corresponding to the maximum membership degree; if the blockage sensitivity coefficient is greater than 1.5, force output of moderate blockage or severe blockage; if the blockage sensitivity coefficient is between 1.2 and 1.5, output according to the maximum membership degree but add a suspected blockage mark.

[0030] Optionally, the L1 level high-frequency flutter self-cleaning control method in step 4 is:

[0031] When the fault level is determined to be mild blockage, the warning indicator light flashes yellow, generating a high-frequency flutter signal. The high-frequency flutter signal is a superposition of a 180 Hz sine wave and Gaussian white noise. The self-cleaning signal is superimposed on the pilot valve's normal control signal. Within 5 seconds after the self-cleaning ends, the signal acquisition, feature extraction, and fault identification steps are re-executed to obtain a new fault level. If the new fault level is normal, the warning is lifted and the self-cleaning is recorded as successful. If it is still mild blockage, the self-cleaning failure counter is incremented. When the number of failures is less than three, the self-cleaning is repeated. When the number of failures reaches three, the fault level is automatically upgraded to moderate blockage, and the corresponding switching control process is initiated.

[0032] Optionally, the control method for automatically switching the backup damping element at level L2 in step 4 is as follows:

[0033] When the fault level is determined to be moderate blockage, the system pressure is first checked. If the system pressure is higher than the preset switching safety pressure threshold, the pressure is reduced to below the switching safety pressure threshold by bypass unloading or load reduction command. A pulse drive signal of a preset duration is sent to the electromagnet of the damping component automatic switching device. The electromagnet drives the shift fork to rotate, pushing the rotary damping component carrier plate to rotate one fixed position, connecting the spare damping component to the oil circuit, and removing the blocked damping component. The feedback signal from the position sensor installed on the carrier plate is read. If no position signal is received within the preset waiting time, the switching device is determined to be faulty, and the process is directly switched to severe blockage shutdown. After the switching is completed, a stabilization period is waited for, during which signals are continuously collected and the blockage sensitivity coefficient is calculated. If the blockage sensitivity coefficient drops below the normal threshold and the fault level drops to normal or mild within the stabilization period, the current state is maintained, the identification information of the original blocked damping component is written into the fault history, and a prompt message is issued. If the blockage sensitivity coefficient is still higher than the normal threshold or the fault level has not dropped below moderate after the stabilization period, the process is upgraded to severe blockage shutdown.

[0034] Optionally, the control method for L3 level safety shutdown in step 4 is as follows: immediately output a cut-off signal through the main controller to disconnect the main pump power source, requiring the cut-off to be completed in a very short time; simultaneously output an opening signal to the bypass unloading valve to rapidly unload the system pressure in a short time; send a fault latch signal to the equipment main controller, the fault latch signal is a holding type and will not be lost when power is off, and prohibit the equipment from starting; display a serious fault prompt on the human-machine interface, and record the current time, fault level and characteristic data snapshot; after manually troubleshooting the fault, clear the fault latch signal through a reset command.

[0035] Optionally, the self-learning update of the benchmark feature library in step 5 is as follows: several effective feature vectors and their corresponding confirmation labels before the occurrence of the fault are used as a new sample set; an incremental single-class support vector machine is used for learning, with the kernel function being the radial basis function; the distance metric between each new sample and the existing fault cluster center of each class is calculated; if there is a class such that the distance is less than a preset multiple of the dispersion of the corresponding class of samples, the new sample is merged into the corresponding class, and the cluster center and covariance matrix of the corresponding class are recalculated; if the above distances of all classes are greater than or equal to the preset multiple, a new fault subclass is created, a new cluster center is assigned, and the corresponding fault label is associated; after a certain number of manual maintenance confirmations are accumulated, the benchmark feature library is fully re-clustered to optimize the classification model.

[0036] Optionally, the cross-correlation time delay in the damping orifice clogging sensitivity coefficient is solved using frequency domain acceleration, including the following sub-steps:

[0037] Step a: Perform zero-mean processing on the main valve core displacement signal and the pilot valve front chamber pressure signal respectively;

[0038] Step b: Calculate the cross power spectrum of the two signals, that is, perform Fourier transform on the two signals respectively, and then multiply the Fourier transform of the main valve core displacement with the conjugate of the Fourier transform of the pilot pressure.

[0039] Step c: Perform an inverse Fourier transform on the cross-power spectrum to obtain the cross-correlation function;

[0040] Step d: Search for the maximum value of the cross-correlation function within the preset delay time range. The delay time corresponding to the maximum value is the required cross-correlation time delay.

[0041] In step e, if the calculated delay time exceeds the normal range, the moving average of historical valid values ​​is used instead, and the current calculation is marked as unreliable.

[0042] A pilot-operated relief valve fault analysis and safety interlock control system is provided to implement a fault analysis method for pilot-operated relief valves. The system includes a signal acquisition unit, an active excitation unit, a feature extraction and diagnosis unit, a hierarchical interlock execution unit, a main controller, and a human-machine interface. The signal acquisition unit includes a first pressure sensor installed at the relief valve inlet, a second pressure sensor installed in the pilot valve front chamber, and a displacement sensor installed on the main valve core, used to synchronize the relief valve inlet pressure signal, the pilot valve front chamber pressure signal, and the main valve core displacement signal at a set sampling frequency. The active excitation unit is used to superimpose the conventional control signal and the excitation signal in real time. The excitation signal includes an active pulse diagnostic excitation signal and a high-frequency chatter self-cleaning signal.

[0043] The feature extraction and diagnosis unit integrates a wavelet packet decomposition module, a clustering analysis coprocessor, and a non-volatile memory. The non-volatile memory is used to store the benchmark feature library. The feature extraction and diagnosis unit is used to extract the feature vector of the real-time signal, perform improved fuzzy C-means clustering calculation, and output the fault type, blockage level, and damping orifice blockage sensitivity coefficient.

[0044] The graded interlocking execution unit includes an early warning indication module, a high-frequency flutter signal generation module, an electromagnetic drive circuit for an automatic damping component switching device, a main pump power source cutoff module, a bypass unloading valve drive module, and a fault latching module.

[0045] The main controller is an industrial logic controller that communicates bidirectionally with each unit. Its built-in logic program is used to implement the timing control and safety interlocking logic of steps 2 to 4.

[0046] The human-computer interaction interface is used to display the fault level, sensitivity coefficient, and historical event records, and to receive input of manual confirmation of maintenance tags to trigger the self-learning update process.

[0047] In summary, the present invention has at least one of the following beneficial technical effects:

[0048] This invention provides a method and system for fault analysis of pilot-operated relief valves, constructs a benchmark feature library for multiple operating conditions and fault types, integrates multi-dimensional features in the time domain, frequency domain, and time-frequency domain, and combines wavelet packet energy spectrum features to accurately distinguish different fault types. It solves the problems of low fault identification and difficulty in distinguishing similar faults in traditional technologies, and significantly improves the identification accuracy of latent faults and early faults.

[0049] By introducing an adaptive active pulse diagnostic excitation mechanism, the valve core dynamic response can be actively stimulated, overcoming the technical bottleneck of weak fault characteristics and insufficient signal identification under natural operating conditions. At the same time, the excitation parameters are adaptively adjusted according to the real-time operating conditions to ensure the stability and effectiveness of the diagnostic process and avoid ineffective diagnostic operations.

[0050] An improved fuzzy C-means clustering algorithm combined with a damping orifice blockage sensitivity coefficient is used to jointly determine faults. The clustering effect is optimized through an intra-class variance penalty term. Relying on multi-dimensional quantitative indicators, accurate fault type and severity determination is achieved, effectively reducing the probability of false positives and false negatives. A tiered safety interlock control logic is also implemented, employing a tiered handling mode of self-cleaning and self-healing, backup damping component switching, and safe shutdown to specifically adapt to different fault levels. This eliminates minor faults, repairs moderate faults, maximizes continuous equipment operation, reduces production losses caused by unnecessary shutdowns, and allows for rapid pressure relief and equipment locking in case of severe faults, preventing major safety accidents.

[0051] Incremental support vector machines are introduced to enable self-learning and updating of the feature library, adapting to changes such as equipment aging and iterative operating conditions, continuously optimizing the fault diagnosis model, and ensuring the diagnostic accuracy and adaptability of the system in long-term operation. The entire system achieves closed-loop control of the entire process, including real-time monitoring of the relief valve status, active diagnosis, intelligent identification, hierarchical handling, and autonomous optimization, effectively reducing equipment failure rate and manual maintenance costs, and improving the stability, safety, and intelligent operation and maintenance level of the hydraulic system. Attached Figure Description

[0052] Figure 1 This is a flowchart illustrating the pilot-operated relief valve fault analysis method of the present invention.

[0053] Figure 2 This is a schematic diagram of the architecture of the pilot-operated relief valve fault analysis and safety interlock control system of the present invention; Detailed Implementation

[0054] The present invention will be further described in detail below with reference to the accompanying drawings.

[0055] This invention discloses a method and system for analyzing the failure of a pilot-operated relief valve.

[0056] Reference Figure 1 and Figure 2 Example 1, a fault analysis method for a pilot-operated relief valve, includes the following steps:

[0057] Step 1: Construct a multi-condition dynamic response benchmark feature library. Through simulation and actual measurement, obtain the time domain, frequency domain, and time-frequency domain features of the overflow valve inlet pressure, pilot valve front chamber pressure, and main valve core displacement signal under health conditions and different degrees of blockage, spring fatigue, and valve core jamming faults, and form a feature vector library with fault labels.

[0058] Step 2: Real-time acquisition of the overflow valve inlet pressure signal, the pilot valve front chamber pressure signal, and the main valve core displacement signal; and periodically or event-triggered superimposed an active pulse diagnostic excitation signal onto the pilot valve control signal to stimulate the valve core dynamic response.

[0059] Step 3: Extract the feature vector of the real-time response signal, use the improved fuzzy C-means clustering algorithm to calculate the Mahalanobis distance between it and each cluster center in the benchmark feature library, and combine it with the damping orifice blockage sensitivity coefficient to perform fault mode identification, and output the fault type and blockage level.

[0060] Step 4: Implement graded safety interlock control according to the fault level, where L1 is high-frequency flutter self-cleaning, L2 is automatic switching to backup damping components, and L3 is safe shutdown.

[0061] Step 5: Use an incremental support vector machine to perform self-learning updates on the benchmark feature library.

[0062] By adopting the above technical solution, firstly, three types of signals—the overflow valve inlet pressure, the pilot valve front chamber pressure, and the main valve core displacement—are simultaneously acquired under healthy conditions and various fault conditions using simulation software such as AMESim and a physical test bench. For each operating condition, time-domain statistical features such as peak value, variance, and waveform factor, frequency-domain features such as dominant frequency offset and power spectral entropy, and time-frequency-domain features such as wavelet packet energy spectrum are extracted to form a feature vector library with fault labels, serving as a benchmark. During online operation, the above three types of signals are acquired in real time at a sampling rate of no less than 5 kHz. Periodically, or when the pressure deviation exceeds a set value of 5%, a short-duration bidirectional pulse excitation is actively superimposed on the pilot valve control signal to stimulate the valve core to generate a measurable dynamic response. After extracting the feature vectors within the real-time window, an improved fuzzy C-means clustering algorithm is used to calculate the Mahalanobis distance with each cluster center in the benchmark library. Combined with the damping orifice blockage sensitivity coefficient, fault mode identification is performed, outputting four levels: normal, mild blockage, moderate blockage, or severe blockage. Based on the identified level of blockage, graded safety interlocks are implemented: high-frequency flutter self-cleaning is initiated for mild blockage, automatic switching to backup damping components is performed for moderate blockage, and a safe shutdown is executed for severe blockage. Finally, the baseline library is continuously updated using manually verified fault samples through incremental support vector machines, allowing the diagnostic model to adapt to valve body aging. This embodiment achieves a fully closed-loop control from offline modeling to online diagnosis, self-cleaning, fault-tolerant switching, and self-learning.

[0063] In Example 2, in step 1, the wavelet packet energy spectrum of the acquired pilot valve pre-cavity pressure signal is decomposed into three layers using the Daubechies 4 wavelet basis. The energy proportion of the eight frequency band nodes in the third layer is extracted as the feature vector. The reconstructed signal energy of each frequency band node is calculated, the total energy is calculated, and the energy proportion of each node is extracted. The energy proportions of three nodes corresponding to the frequency ranges of 100Hz-200Hz, 200Hz-400Hz, and 400Hz-800Hz are selected as the feature vectors. This is used to distinguish between damping orifice blockage and spring fatigue.

[0064] By adopting the above technical solution, the Daubechies4 wavelet is selected as the mother wavelet to perform three-level wavelet packet decomposition on the pilot valve pre-cavity pressure signal. The third level generates eight frequency band nodes, each with a frequency width equal to one-sixteenth of the sampling frequency. For each node, the energy of the node is obtained by reconstructing its time-domain signal and summing the squares of the amplitudes of each sampling point of the reconstructed signal. The total energy is obtained by summing the energies of all eight nodes, and then the percentage of energy of each node in the total energy is calculated. Experimental analysis shows that damping orifice blockage mainly affects the energy distribution in three frequency bands: 100-200 Hz, 200-400 Hz, and 400-800 Hz, while spring fatigue has a different impact frequency band. Therefore, the energy percentages of these three specific nodes are extracted and combined into a feature sub-vector. When the sub-vector extracted online in real time is closer to the cluster center of the blockage category in the benchmark library, it is determined to be damping orifice blockage; if it is closer to the spring fatigue category, it is determined to be spring fatigue. This feature effectively solves the problem that a single pressure threshold cannot distinguish between the two types of faults.

[0065] Example 3, the generation and adaptive adjustment of the active pulse diagnostic excitation signal in step 2 includes the following sub-steps:

[0066] Step 21: Set the basic excitation waveform to a bidirectional rectangular pulse sequence, with each pulse width of 0.05 seconds, a total of four pulses, a pulse interval of 0.025 seconds, and a total excitation duration of no more than 0.3 seconds;

[0067] Step 22: Obtain the steady-state value of the main valve core displacement in real time and compare it with the reference displacement corresponding to the rated flow rate;

[0068] Step 23: Adaptively adjust the excitation amplitude based on the ratio of steady-state displacement to reference displacement: when the ratio is greater than or equal to the set maximum flow rate, the excitation amplitude is set to 5% of the rated maximum control signal value; when the ratio is less than or equal to the set minimum flow rate, the excitation amplitude is set to 20%; otherwise, it is set to 8%.

[0069] Step 24: The generated excitation signal is superimposed on the conventional control signal of the pilot valve, and the superposition method is bidirectional alternating positive and negative pulses;

[0070] Step 25: Within 0.1 seconds after the excitation ends, monitor the maximum deviation of the inlet pressure. If the maximum deviation exceeds 10% of the set pressure, mark the current excitation as invalid and automatically reduce the amplitude of the next excitation by 20%.

[0071] By adopting the above technical solution, the basic excitation waveform is set as a bidirectional alternating positive and negative rectangular pulse sequence, with each pulse width of 0.05 seconds, a total of four pulses, a pulse interval of 0.025 seconds, and a total duration not exceeding 0.3 seconds, to avoid excessive disturbance to the system pressure. The system monitors the steady-state value of the main valve core displacement in real time and compares it with the reference displacement corresponding to the rated flow rate. When the ratio of the steady-state displacement to the reference displacement is greater than or equal to 0.8, it indicates that the system is in a high-flow-rate condition. At this time, the excitation amplitude is set to 5% of the maximum value of the rated control signal to prevent pressure overshoot caused by superimposed signals. When the ratio is less than or equal to 0.2, it is a low-flow-rate condition, and the valve core response is weak. The excitation amplitude is increased to 20% to enhance the visibility of the diagnostic signal. For other intermediate conditions, it is set to 8%. The excitation signal is superimposed on the conventional control signal in a bidirectional alternating manner. The transient pressure difference generated by the positive and negative pulses forces the valve core to make micro-movements, thereby exposing the response delay caused by blockage. Within 0.1 seconds after the excitation ends, the system monitors the maximum deviation of the inlet pressure from the set pressure. If the deviation exceeds 10%, it indicates that the excitation has caused unacceptable interference to the main system. At this time, the current excitation is marked as invalid, and the amplitude of the next excitation will be automatically reduced by 20% to maintain diagnostic capability while ensuring safety.

[0072] Example 4, the improved fuzzy C-means clustering algorithm and the calculation of the damping orifice blockage sensitivity coefficient in step 3 include the following sub-steps:

[0073] Step 31: Bandpass filtering and normalization are performed on the inlet pressure, pilot valve front chamber pressure and main valve core displacement signals in the real-time acquisition window to make the dimensions of each characteristic consistent.

[0074] Step 32: Extract the time-domain, frequency-domain, and time-frequency-domain features to form a real-time feature vector;

[0075] Step 33: Calculate the Mahalanobis distance between the real-time feature vector and the fault cluster center of each class in the benchmark feature library. The Mahalanobis distance is calculated by weighting the covariance matrix of the corresponding class samples.

[0076] Step 34: Calculate the fuzzy membership degree of the real-time feature vector to each fault type based on Mahalanobis distance. The membership degree function adopts the inverse square weighted form.

[0077] Step 35: Construct an improved objective function. The objective function adds a penalty term to the original fuzzy C-means clustering objective. The penalty term is the sum of the inverses of the intra-class variances of each fault sample in the two feature dimensions of damping orifice blockage sensitivity coefficient and pilot pressure fluctuation peak-to-peak value, multiplied by the penalty coefficient. The cluster centers are iteratively updated using the gradient descent method until convergence.

[0078] Step 36, calculate the damping orifice clogging sensitivity coefficient. ;

[0079] ;

[0080] in This represents the time delay corresponding to the maximum peak value of the cross-correlation function between the pilot valve inlet pressure and the main valve spool displacement. The rise time of the step response of the main valve core displacement. The standard deviation of the leading pressure, This represents the average value of the pilot pressure.

[0081] Step 37: Determine the fault level based on the maximum membership degree and the blockage sensitivity coefficient: If the maximum membership degree is greater than or equal to 0.6 and the blockage sensitivity coefficient is less than or equal to 1.2, output the fault level corresponding to the maximum membership degree; if the blockage sensitivity coefficient is greater than 1.5, force output of moderate blockage or severe blockage; if the blockage sensitivity coefficient is between 1.2 and 1.5, output according to the maximum membership degree but add a suspected blockage mark.

[0082] By adopting the above technical solution

[0083] First, the inlet pressure, pilot pressure, and main valve core displacement signals within the real-time acquisition window are bandpass filtered, with a passband range of 10 Hz to 1 kHz to filter out low-frequency drift and high-frequency noise. Then, each feature quantity is normalized to eliminate the influence of dimensions. Next, the twelve-dimensional feature vector described in Example 1 is extracted. The Mahalanobis distance between this real-time vector and the cluster center of each fault type in the benchmark library is calculated. The Mahalanobis distance is weighted using the covariance matrix of the samples in that class, which can eliminate the influence of inter-feature correlation on the distance metric. Fuzzy membership is calculated based on the Mahalanobis distance, using an inverse square weighted form, where the fuzziness exponent is 2. A penalty term is added to the objective function of traditional fuzzy C-means clustering. This penalty term is the sum of the inverses of the intra-class variances of each fault sample in the two feature dimensions of damping orifice blockage sensitivity coefficient and pilot pressure fluctuation peak-to-peak value, multiplied by a penalty coefficient of 0.1. This penalty term forces the clustering process to pay more attention to the two feature dimensions closely related to blockage, improving the sensitivity of blockage identification. The cluster centers are iteratively updated using gradient descent until the change in the objective function is less than one ten-thousandth. The damping orifice clogging sensitivity coefficient is calculated using the following formula:

[0084] in This represents the time delay corresponding to the maximum peak value of the cross-correlation function between the pilot valve inlet pressure and the main valve spool displacement. The rise time of the step response of the main valve core displacement. The standard deviation of the leading pressure, This is the average of the pilot pressure. Finally, a joint determination of the fault level is made: if the maximum membership degree is ≥0.6 and... If ≤1.2, output the level corresponding to the highest membership degree; if >1.5, forced output moderate or severe blockage; if Between 1.2 and 1.5, the output shows the highest membership level but with an additional suspected blockage flag to alert operators.

[0085] Example 5, the L1 level high-frequency flutter self-cleaning control method in step 4 is:

[0086] When the fault level is determined to be mild blockage, the warning indicator light flashes yellow, generating a high-frequency flutter signal. The high-frequency flutter signal is a superposition of a 180 Hz sine wave and Gaussian white noise. The self-cleaning signal is superimposed on the pilot valve's normal control signal. Within 5 seconds after the self-cleaning ends, the signal acquisition, feature extraction, and fault identification steps are re-executed to obtain a new fault level. If the new fault level is normal, the warning is lifted and the self-cleaning is recorded as successful. If it is still mild blockage, the self-cleaning failure counter is incremented. When the number of failures is less than three, the self-cleaning is repeated. When the number of failures reaches three, the fault level is automatically upgraded to moderate blockage, and the corresponding switching control process is initiated.

[0087] By adopting the above technical solution, when the system determines the fault level to be mild blockage, it first triggers a warning indicator to flash yellow and displays a self-cleaning in progress message on the human-machine interface. Subsequently, a high-frequency flutter signal is generated. The core of this signal is a 180 Hz sine wave with an amplitude set to 4% of the maximum value of the rated control signal. The superimposed auxiliary component is Gaussian white noise with a root mean square value set to 20% of the sine wave amplitude. The white noise prevents impurities from resonating and adhering under single-frequency vibration, improving the cleaning effect. This signal is superimposed on the pilot valve's conventional control signal and applied continuously for 60 seconds. During this period, the valve core generates micron-level reciprocating motion under high-frequency, small-amplitude vibration, causing localized turbulence in the oil, loosening impurities adhering to the damping orifice wall and washing them away with the oil. After self-cleaning, wait 5 seconds to allow the system to return to a relatively steady state, and then re-execute the signal acquisition, feature extraction, and fault identification steps. If the newly identified fault level returns to normal, the warning is cleared and a successful self-cleaning event is recorded. If the blockage remains mild, the self-cleaning failure counter is incremented. When the number of failures is less than 3, the self-cleaning process is repeated. If three consecutive self-cleaning attempts fail, the system determines that self-cleaning cannot resolve the blockage, automatically upgrades the fault level to moderate blockage, and proceeds to the automatic switching of backup damping components process described in Example 6.

[0088] Example 6, the control method for automatic switching of the backup damping element at level L2 in step 4 is as follows:

[0089] When the fault level is determined to be moderate blockage, the system pressure is first checked. If the system pressure is higher than the preset switching safety pressure threshold, the pressure is reduced to below the switching safety pressure threshold by bypass unloading or load reduction command. A pulse drive signal of a preset duration is sent to the electromagnet of the damping component automatic switching device. The electromagnet drives the shift fork to rotate, pushing the rotary damping component carrier plate to rotate one fixed position, connecting the spare damping component to the oil circuit, and removing the blocked damping component. The feedback signal from the position sensor installed on the carrier plate is read. If no position signal is received within the preset waiting time, the switching device is determined to be faulty, and the process is directly switched to severe blockage shutdown. After the switching is completed, a stabilization period is waited for, during which signals are continuously collected and the blockage sensitivity coefficient is calculated. If the blockage sensitivity coefficient drops below the normal threshold and the fault level drops to normal or mild within the stabilization period, the current state is maintained, the identification information of the original blocked damping component is written into the fault history, and a prompt message is issued. If the blockage sensitivity coefficient is still higher than the normal threshold or the fault level has not dropped below moderate after the stabilization period, the process is upgraded to severe blockage shutdown.

[0090] By adopting the above technical solution, when the fault level rises to moderate blockage, the system first detects the current system pressure. If the actual pressure is higher than the preset switching safety pressure threshold, which is generally set to 0.6 times the rated pressure, the system reduces the load or opens the bypass unloading valve to lower the pressure below the threshold to ensure that the switching mechanism is not damaged by high-pressure impact during the switching process. Subsequently, a pulse drive signal with a preset duration is sent to the electromagnet of the damping component automatic switching device. This duration is sufficient to drive the shift fork to complete one rotation. After the electromagnet is energized, it drives the shift fork to move, pushing the rotary damping component carrier plate to rotate to a fixed position, usually 60 degrees, so that the spare damping component accurately enters the oil circuit position, while removing the blocked damping component from the oil circuit. A Hall position sensor is installed at the bottom of the carrier plate. After sending the drive pulse, the system waits for a preset time, such as 500 milliseconds. If no position signal is received from the sensor within this time, it is considered that the switching device has mechanically jammed or the electromagnet has failed. At this time, the system does not continue to attempt switching and directly proceeds to the severe blockage shutdown process of Example 7. If the switch is successful, the system waits for a stabilization period, typically 30 seconds, to allow the oil pressure and flow to rebalance. During this period, signals are continuously collected and the blockage sensitivity coefficient is calculated. If the blockage sensitivity coefficient drops below the normal threshold of 1.2 and the fault level decreases to normal or mild within the stabilization period, the current state is maintained. The unique identifier of the original blockage damping component, such as the workstation number and usage duration, is written into the fault history in the non-volatile memory, and a prompt message is issued through the human-machine interface to remind the operator to replace the blockage damping component during the next planned shutdown. If the blockage sensitivity coefficient is still higher than 1.2 or the fault level has not decreased to below moderate after the stabilization period, it indicates that the backup damping component may also be blocked or the switch failed to completely resolve the problem. In this case, the system is escalated to a severe blockage shutdown procedure.

[0091] In Example 7, the control method for L3 level safety shutdown in step 4 is as follows: immediately output a cut-off signal through the main controller to disconnect the main pump power source, requiring the cut-off to be completed in a very short time; simultaneously output an opening signal to the bypass unloading valve to rapidly unload the system pressure in a short time; send a fault latch signal to the equipment main controller, the fault latch signal is a holding type and will not be lost when power is off, and prohibit the equipment from starting; display a serious fault prompt on the human-machine interface, and record the current time, fault level and characteristic data snapshot; after manually troubleshooting the fault, clear the fault latch signal through a reset command.

[0092] By adopting the above technical solution, when a severe blockage or jamming is detected, the system immediately activates the highest priority safety protection program. The main controller first outputs a cut-off signal to the contactor coil of the main pump motor, requiring the contactor to complete the disconnection action within a very short time, such as 50 milliseconds, thereby cutting off the power source of the hydraulic system. Simultaneously, a parallel opening signal is output to the solenoid coil of the bypass unloading valve, causing the unloading valve to quickly open. The system pressure is rapidly released through the unloading channel, required to drop to below 0.2 times the rated pressure within a short time, such as 100 milliseconds, to prevent high-pressure oil from damaging other components in a blocked state. Subsequently, the main controller sends a fault latch signal to the equipment's host controller. This signal has a holding characteristic and will not be lost even if the system is powered off. It also prevents the equipment's start-up relay from engaging, ensuring that the system cannot be restarted until the fault is resolved. The human-machine interface switches to a severe fault warning screen, displaying specific information and automatically recording the current time, fault level, and a snapshot of characteristic data before the fault occurred for easy post-fault analysis. Once the operator arrives, they will manually inspect and replace the blocked damping components. After confirming that the fault has been eliminated, a reset command will be sent via a rotary reset switch or a host computer to clear the fault latch signal. Only then can the equipment be restarted.

[0093] In Example 8, the method for updating the baseline feature library through self-learning in step 5 is as follows: several effective feature vectors and their corresponding confirmation labels before the occurrence of the current fault are used as a new sample set; an incremental single-class support vector machine is used for learning, with the kernel function being the radial basis function; the distance metric between each new sample and the existing fault cluster center of each class is calculated; if there exists a class such that the distance is less than a preset multiple of the dispersion of the corresponding class of samples, the new sample is merged into the corresponding class, and the cluster center and covariance matrix of the corresponding class are recalculated; if the above distances of all classes are greater than or equal to the preset multiple, a new fault subclass is created, a new cluster center is assigned, and the corresponding fault label is associated; after a certain number of manual maintenance confirmations are accumulated, the baseline feature library is fully re-clustered to optimize the classification model.

[0094] By adopting the above technical solution, the self-learning process is only triggered when the operator confirms the fault label through the human-machine interface after manual maintenance, thus avoiding the contamination of the benchmark database by erroneous samples. After the operator confirms the actual fault type, the system uses the last ten valid feature vectors stored before the fault occurred and their corresponding confirmation labels as a new sample set. An incremental single-class support vector machine is used for learning, with a radial basis function selected as the kernel function and a kernel width parameter of 0.1. For each new sample, the Mahalanobis distance between it and the existing fault cluster centers of each class is calculated. If there exists a class such that the distance is less than 0.3 times the intra-class standard deviation of the samples in that class, the new sample is considered to belong to that class, and the samples are merged, and the cluster centers and covariance matrix of that class are recalculated. If the distances of all classes are greater than or equal to 0.3 times the intra-class standard deviation, it indicates that the sample represents a fault mode not yet covered by the model. In this case, a new fault subclass is created, a new cluster center is assigned, and the new sample is used as the initial sample of that subclass. To prevent the model from growing indefinitely, after every 100 manual maintenance confirmations, the system automatically re-clusters the entire baseline feature library using the standard fuzzy C-means algorithm. The initial number of clusters is set to the current number of categories plus two. Then, the clustering quality is evaluated based on the Davidson-Bolding index to select the optimal number of categories. This mechanism enables the diagnostic model to adapt to long-term drift factors such as valve body wear and changes in oil contamination levels.

[0095] Example 9: The cross-correlation time delay in the damping orifice blockage sensitivity coefficient is solved using frequency domain acceleration, including the following sub-steps:

[0096] Step a: Perform zero-mean processing on the main valve core displacement signal and the pilot valve front chamber pressure signal respectively;

[0097] Step b: Calculate the cross power spectrum of the two signals, that is, perform Fourier transform on the two signals respectively, and then multiply the Fourier transform of the main valve core displacement with the conjugate of the Fourier transform of the pilot pressure.

[0098] Step c: Perform an inverse Fourier transform on the cross-power spectrum to obtain the cross-correlation function;

[0099] Step d: Search for the maximum value of the cross-correlation function within the preset delay time range. The delay time corresponding to the maximum value is the required cross-correlation time delay.

[0100] In step e, if the calculated delay time exceeds the normal range, the moving average of historical valid values ​​is used instead, and the current calculation is marked as unreliable.

[0101] By adopting the above technical solution, a frequency domain method is used instead of a time domain sliding correlation to reduce online calculation latency. Specifically, the mean values ​​of the main valve core displacement signal and the pilot valve pre-chamber pressure signal are first subtracted to achieve zero-mean processing, thus eliminating the influence of the DC component on the correlation. Then, a Fast Fourier Transform (FFT) is performed on each signal to obtain their respective spectra. The cross-power spectrum is calculated by multiplying the Fourier transform result of the main valve core displacement signal point-by-point with the conjugate complex number of the Fourier transform result of the pilot pressure signal. An Inverse Fourier Transform is performed on the cross-power spectrum to obtain the cross-correlation function, which is defined on the positive and negative time delay axes. Within a preset delay time range, typically ±0.1 seconds, the system searches for the maximum value of the cross-correlation function; the delay time corresponding to this maximum value is the required value. Because the frequency domain method leverages the parallel computation advantage of the Fast Fourier Transform, its computation speed is much faster than that of the time domain sliding correlation. To prevent outlier interference, an additional verification step is added: if the calculated... If the time exceeds the normal physical range, such as less than 0.005 seconds or greater than 0.05 seconds, the calculation result is discarded and replaced with the moving average of the three most recent valid historical values. The calculation is then marked as unreliable, and the credibility of this feature is appropriately reduced in subsequent fault determination.

[0102] Example 10: A pilot-operated relief valve fault analysis and safety interlock control system, used to implement a pilot-operated relief valve fault analysis method. The system includes a signal acquisition unit, an active excitation unit, a feature extraction and diagnosis unit, a hierarchical interlock execution unit, a main controller, and a human-machine interface. The signal acquisition unit includes a first pressure sensor installed at the relief valve inlet, a second pressure sensor installed in the pilot valve front chamber, and a displacement sensor installed on the main valve core, used to synchronize the relief valve inlet pressure signal, the pilot valve front chamber pressure signal, and the main valve core displacement signal at a set sampling frequency. The active excitation unit is used to superimpose the conventional control signal and the excitation signal in real time. The excitation signal includes an active pulse diagnostic excitation signal and a high-frequency chatter self-cleaning signal.

[0103] The feature extraction and diagnosis unit integrates a wavelet packet decomposition module, a clustering analysis coprocessor, and a non-volatile memory. The non-volatile memory is used to store the benchmark feature library. The feature extraction and diagnosis unit is used to extract the feature vector of the real-time signal, perform improved fuzzy C-means clustering calculation, and output the fault type, blockage level, and damping orifice blockage sensitivity coefficient.

[0104] The graded interlocking execution unit includes an early warning indication module, a high-frequency flutter signal generation module, an electromagnetic drive circuit for an automatic damping component switching device, a main pump power source cutoff module, a bypass unloading valve drive module, and a fault latching module.

[0105] The main controller is an industrial logic controller that communicates bidirectionally with each unit. Its built-in logic program is used to implement the timing control and safety interlocking logic of steps 2 to 4.

[0106] The human-computer interaction interface is used to display the fault level, sensitivity coefficient, and historical event records, and to receive input of manual confirmation of maintenance tags to trigger the self-learning update process.

[0107] The following specific embodiments illustrate the implementation principle of the present invention:

[0108] Taking the hydraulic system of a certain engineering machinery as an application scenario, the main safety valve of this hydraulic system adopts a pilot-operated relief valve, and this relief valve has been modified according to the structure of patent 201721160956.3, equipped with a rotatable damping component carrier plate. Two replaceable damping components, namely damping component A and damping component B, are installed on the carrier plate. The rated working pressure of the system is 25 MPa, and the rated flow rate is 200 liters per minute. The sampling frequency of the control system is set to 10 kHz, the main controller is a Siemens S7-1200 series programmable logic controller, the feature extraction and diagnostic unit uses a Zynq series field-programmable gate array chip, and the human-machine interface is a 7-inch industrial touch screen.

[0109] Before the equipment is put into operation, step 1 of Example 1 is performed first. Researchers establish an accurate model of the pilot-operated relief valve in the AMESim simulation platform. The model includes parameters such as the main valve core mass, pilot valve pressure regulating spring stiffness, main valve spring stiffness, and damping orifice diameter. Simultaneously, the same relief valve is installed on an actual test bench, and different faults are artificially created to obtain measured data. Simulation and actual testing cover seven operating conditions: healthy state, a 10% reduction in the effective area of ​​the damping orifice representing mild blockage, a 40% reduction representing moderate blockage, an 80% reduction representing severe blockage, a 10% decrease in pilot valve spring stiffness representing spring fatigue, a 1.5-fold increase in the additional friction force of the main valve core representing jamming, and complete blockage of the damping orifice. For each operating condition, the relief valve inlet pressure, pilot valve front chamber pressure, and main valve core displacement signal are collected for 60 seconds each. Following the method in Example 2, the pilot valve pre-chamber pressure signal was decomposed into three layers of wavelet packets using Daubechies 4 wavelets. The energy proportions of the eight frequency band nodes in the third layer were extracted, and the energy proportions of three nodes (100-200 Hz, 200-400 Hz, and 400-800 Hz) were selected as wavelet packet energy spectrum features. Simultaneously, peak value, variance, and waveform factor in the time domain, dominant frequency offset and power spectral entropy in the frequency domain, and other features in the time and frequency domains were extracted to form a twelve-dimensional feature vector. Healthy state was labeled L0, mild blockage L1, moderate blockage L2, severe blockage and jamming L3, and spring fatigue was separately labeled as other faults but not the main focus of this case. These feature vectors were clustered using an improved fuzzy C-means clustering algorithm, resulting in four cluster centers, which were stored in non-volatile flash memory, completing the construction of the benchmark feature library.

[0110] During normal operation, the signal acquisition unit synchronously acquires signals from three sensors at a sampling frequency of 10 kHz. The first pressure sensor is installed at the inlet of the relief valve, with a range of 0 to 40 MPa, and outputs a current signal of 4 to 20 mA. The second pressure sensor is installed in the front chamber of the pilot valve, with a range of 0 to 10 MPa. The eddy current displacement sensor is installed at the end of the main valve core, with a range of 0 to 5 mm, and outputs a voltage signal. The signal conditioning circuit converts the sensor signals into a voltage of 0 to 10 volts, which is then sent to the feature extraction and diagnostic unit via a 16-bit analog-to-digital converter.

[0111] An internal timer is set in the main controller to automatically trigger an active pulse diagnostic excitation every 300 seconds. Simultaneously, an event trigger condition is set: when the absolute value of the deviation between the inlet pressure and the set pressure exceeds 1.25 MPa (5% of the set pressure) and lasts for more than 1 second, excitation is also triggered. Assuming the system is under medium flow conditions at a certain moment, the steady-state displacement of the main valve core is 1.2 mm, while the reference displacement corresponding to the rated flow is 1.5 mm, a ratio of 0.8, which falls within the boundary of high flow conditions. Following steps 21 to 23 of Example 3, the basic excitation waveform is a bidirectional rectangular pulse sequence, each pulse width 0.05 seconds, for a total of four pulses, with a pulse interval of 0.025 seconds. Since the ratio is equal to 0.8, the excitation amplitude is set to 5% of the maximum value of the rated control signal. The maximum value of this rated control signal is 10 volts, therefore the excitation amplitude is 0.5 volts. Step 24 superimposes the generated excitation signal onto the conventional control signal of the pilot valve, using alternating positive and negative superposition. Within 0.1 seconds after the excitation ended, the maximum deviation of the monitored inlet pressure was 0.3 MPa, which did not exceed 10% of the set pressure of 25 MPa, i.e., 2.5 MPa. Therefore, the excitation was valid. The active excitation caused the valve core to produce a small additional displacement, and the waveform of this displacement response was completely recorded.

[0112] The feature extraction and diagnosis unit extracts real-time feature vectors within the excitation window, obtaining twelve-dimensional values. Then, fault identification is performed according to steps 31 to 37 of Example 4. First, the inlet pressure, pilot pressure, and main valve core displacement signals are bandpass filtered from 10 Hz to 1 kHz, and then normalized. The Mahalanobis distances between this real-time feature vector and the four cluster centers L0, L1, L2, and L3 in the benchmark library are calculated, yielding values ​​of 1.2, 0.5, 2.1, and 3.8, respectively. Fuzzy membership degrees are calculated based on the Mahalanobis distances, resulting in a membership degree of 0.15 for L0, 0.75 for L1, 0.09 for L2, and 0.01 for L3. Simultaneously, the damping orifice blockage sensitivity coefficient is calculated. First, the main valve core displacement and pilot pressure signals are zero-mean processed. Then, the frequency domain acceleration method of Example 9 is used: a 2048-point Fast Fourier Transform is performed on each signal to calculate the cross-power spectrum, followed by an inverse Fourier Transform to obtain the cross-correlation function. The maximum value is searched within a range of ±0.1 seconds to obtain... The rise time of the step response is 0.012 seconds. The time was measured to be 0.010 seconds by applying a step command. The standard deviation of the pilot pressure was 0.15 MPa, and the mean Ppilot value was 1.2 MPa. Substituting into the formula... = (0.012 / 0.010) × (1 + 2.0 × 0.15 / 1.2) = 1.2 × (1 + 0.25) = 1.5. According to the decision rule in step 37, the maximum membership degree is 0.75, which is greater than 0.6 and... The value is equal to 1.5, greater than 1.2 but not exceeding 1.5, therefore it falls between 1.2 and 1.5. The output is the L1 level corresponding to the highest membership degree, with a suspected congestion flag added. The HMI displays "mild congestion," with a congestion sensitivity coefficient of 1.5.

[0113] The system enters the L1 level high-frequency chatter self-cleaning control described in Example 5. The main controller sends a command to the hierarchical interlocking execution unit, and the warning indicator light flashes yellow once per second. Simultaneously, a high-frequency chatter signal is generated: a 180 Hz sine wave with an amplitude of 0.4 volts, plus Gaussian white noise with a root mean square value of 0.08 volts. This signal is superimposed on the pilot valve control signal through the active excitation unit and lasts for 60 seconds. During this period, the operator sees a notification on the human-machine interface indicating that self-cleaning is in progress. After 60 seconds, the system waits for 5 seconds and then re-executes signal acquisition and feature extraction. The newly calculated feature vector has a Mahalanobis distance of 0.8 from the L0 cluster center, the maximum membership degree reaches 0.85 for L0, and the blockage sensitivity coefficient decreases to 1.1. The system determines that self-cleaning is successful, removes the yellow warning, records a successful event, and resumes normal operation.

[0114] Three months later, the device triggered diagnostics again. This time, the calculated feature vector had a membership degree of 0.82 with L1 and a blockage sensitivity coefficient of 1.6. Following the rules of Example 4, If the value is greater than 1.5, the forced output is a moderate blockage, i.e., L2 level. The system enters the automatic switching process for the backup damper described in Example 6. First, the system pressure is detected to be 24 MPa, which is higher than the switching safety pressure threshold of 15 MPa, i.e., 0.6 times the rated pressure. The main controller reduces the system pressure to 14 MPa through the bypass unloading valve. Then, a 24-volt pulse drive signal with a duration of 300 milliseconds is sent to the electromagnet of the automatic damper switching device. The electromagnet drives the fork to rotate, pushing the rotary damper carrier plate to rotate 60 degrees clockwise, connecting the backup damper B to the oil circuit, and simultaneously removing the blocked damper A. The Hall position sensor installed at the bottom of the carrier plate returns a position signal within 200 milliseconds, indicating a successful switch. The system waits for a 30-second stabilization period, during which the blockage sensitivity coefficient is calculated once per second. At the 10th second... The value dropped to 1.0, then to 0.9 at 20 seconds, and stabilized at 0.85 at 30 seconds, restoring the fault level to L0. The main controller wrote the station number 01 of damper A and its usage time of 3650 hours into the EEPROM fault history. The HMI displayed "Switching successful; please replace the blocked damper during the next shutdown." The operator recorded this information and continued production.

[0115] Six months later, damper B also gradually became clogged. A diagnostic output... The membership level is 2.0, with a maximum membership degree of L3. The system enters the L3 level safety shutdown procedure of Example 7. The main controller immediately outputs a cut-off signal to the main pump motor contactor coil. The contactor disconnects within 40 milliseconds, and the main pump stops supplying oil. Simultaneously, a 24-volt voltage is output to the bypass unloading valve solenoid coil. The unloading valve fully opens within 80 milliseconds, and the system pressure rapidly drops from 25 MPa to 2 MPa. The main controller sends a fault latch signal to the equipment master control. This signal is maintained by a self-holding relay after power failure, and the equipment start-up circuit is disconnected. The HMI displays a flashing red warning: severe blockage or jamming, requiring manual replacement of the damping component and reset. At the same time, the current time, fault level L3, and a snapshot of characteristic data from the 10 seconds prior to the fault are automatically recorded. After the operator arrives, damping component B is replaced, and it is confirmed that there are no other faults. Then, a reset command is sent via the reset button on the HMI to clear the fault latch signal, and the equipment restarts.

[0116] After each manual maintenance, the operator confirms the actual fault label through the human-machine interface. The system performs self-learning according to the method in Example 8. For example, if the actual fault is confirmed to be moderate congestion, the system uses the last ten valid feature vectors and their labels before the fault occurred as new samples and employs incremental single-class support vector machine learning. The Mahalanobis distance between the new samples and the existing L2 cluster centers is calculated to be 0.25 times the within-class standard deviation, which is less than 0.3 times. Therefore, the samples are merged into the L2 class, and the cluster centers are recalculated. Every 100 manual confirmations, the system automatically performs a full re-clustering to optimize the classification model.

[0117] During system operation, the frequency domain acceleration algorithm in Example 9 reduces the time delay calculation time for each cross-correlation to only 2 milliseconds, far lower than the 50 milliseconds for time domain sliding correlation, thus meeting the real-time requirements at a 10 kHz sampling rate. In the system architecture described in Example 10, each unit communicates via a Profinet bus, and the main controller polls for 10 milliseconds, ensuring a rapid response to safety interlocks.

[0118] The above specific implementation examples verify the effectiveness of the pilot-operated relief valve fault diagnosis and safety interlock control method of the present invention. This method can proactively identify early blockages, achieve maintenance-free operation through high-frequency chatter self-cleaning, automatically switch to a backup damping component when self-cleaning is ineffective, and safely shut down the system in case of a serious fault. It also possesses self-learning capabilities, significantly improving the reliability and maintenance efficiency of the hydraulic system.

[0119] The above are all preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Therefore, all equivalent changes made in accordance with the structure, shape and principle of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A fault analysis method for pilot-operated relief valves, characterized in that, Includes the following steps: Step 1: Construct a multi-condition dynamic response benchmark feature library. Through simulation and actual measurement, obtain the time domain, frequency domain, and time-frequency domain features of the overflow valve inlet pressure, pilot valve front chamber pressure, and main valve core displacement signal under health conditions and different degrees of blockage, spring fatigue, and valve core jamming faults, and form a feature vector library with fault labels. Step 2: Real-time acquisition of the overflow valve inlet pressure signal, the pilot valve front chamber pressure signal, and the main valve core displacement signal; and periodically or event-triggered superimposed an active pulse diagnostic excitation signal onto the pilot valve control signal to stimulate the valve core dynamic response. Step 3: Extract the feature vector of the real-time response signal, use the improved fuzzy C-means clustering algorithm to calculate the Mahalanobis distance between it and each cluster center in the benchmark feature library, and combine it with the damping orifice blockage sensitivity coefficient to perform fault mode identification, and output the fault type and blockage level. Step 4: Implement graded safety interlock control according to the fault level, where L1 is high-frequency flutter self-cleaning, L2 is automatic switching to backup damping components, and L3 is safe shutdown. Step 5: Use an incremental support vector machine to perform self-learning updates on the benchmark feature library.

2. The fault analysis method for a pilot-operated relief valve according to claim 1, characterized in that, In step 1, the wavelet packet energy spectrum is decomposed using the Daubechies4 wavelet basis on the acquired pilot valve pre-cavity pressure signal in three layers. The energy proportion of the eight frequency band nodes in the third layer is extracted as the feature vector. The reconstructed signal energy of each frequency band node is calculated, the total energy is calculated, and the energy proportion of each node is extracted. The energy proportions of three nodes corresponding to the frequency ranges of 100Hz-200Hz, 200Hz-400Hz, and 400Hz-800Hz are selected as the feature vectors. This is used to distinguish between damping orifice blockage and spring fatigue.

3. The pilot-operated relief valve fault analysis method according to claim 2, characterized in that, Step 2, the generation and adaptive adjustment of the active pulse diagnostic excitation signal, includes the following sub-steps: Step 21: Set the basic excitation waveform to a bidirectional rectangular pulse sequence, with each pulse width of 0.05 seconds, a total of four pulses, a pulse interval of 0.025 seconds, and a total excitation duration of no more than 0.3 seconds; Step 22: Obtain the steady-state value of the main valve core displacement in real time and compare it with the reference displacement corresponding to the rated flow rate; Step 23: Adaptively adjust the excitation amplitude based on the ratio of steady-state displacement to reference displacement: when the ratio is greater than or equal to the set maximum flow rate, the excitation amplitude is set to 5% of the rated maximum control signal value; when the ratio is less than or equal to the set minimum flow rate, the excitation amplitude is set to 20%; otherwise, it is set to 8%. Step 24: The generated excitation signal is superimposed on the conventional control signal of the pilot valve, and the superposition method is bidirectional alternating positive and negative pulses; Step 25: Within 0.1 seconds after the excitation ends, monitor the maximum deviation of the inlet pressure. If the maximum deviation exceeds 10% of the set pressure, mark the current excitation as invalid and automatically reduce the amplitude of the next excitation by 20%.

4. The pilot-operated relief valve fault analysis method according to claim 3, characterized in that, Step 3, which involves improving the fuzzy C-means clustering algorithm and calculating the damping orifice clogging sensitivity coefficient, includes the following sub-steps: Step 31: Bandpass filtering and normalization are performed on the inlet pressure, pilot valve front chamber pressure and main valve core displacement signals in the real-time acquisition window to make the dimensions of each characteristic consistent. Step 32: Extract the time-domain, frequency-domain, and time-frequency-domain features to form a real-time feature vector; Step 33: Calculate the Mahalanobis distance between the real-time feature vector and the fault cluster center of each class in the benchmark feature library. The Mahalanobis distance is calculated by weighting the covariance matrix of the corresponding class samples. Step 34: Calculate the fuzzy membership degree of the real-time feature vector to each fault type based on Mahalanobis distance. The membership degree function adopts the inverse square weighted form. Step 35: Construct an improved objective function. The objective function adds a penalty term to the original fuzzy C-means clustering objective. The penalty term is the sum of the inverses of the intra-class variances of each fault sample in the two feature dimensions of damping orifice blockage sensitivity coefficient and pilot pressure fluctuation peak-to-peak value, multiplied by the penalty coefficient. The cluster centers are iteratively updated using the gradient descent method until convergence. Step 36, calculate the damping orifice clogging sensitivity coefficient. ; ； in This represents the time delay corresponding to the maximum peak value of the cross-correlation function between the pilot valve inlet pressure and the main valve spool displacement. The rise time of the step response of the main valve core displacement. The standard deviation of the leading pressure, This represents the average value of the pilot pressure. Step 37: Determine the fault level based on the maximum membership degree and the blockage sensitivity coefficient: If the maximum membership degree is greater than or equal to 0.6 and the blockage sensitivity coefficient is less than or equal to 1.2, output the fault level corresponding to the maximum membership degree; if the blockage sensitivity coefficient is greater than 1.5, force output of moderate blockage or severe blockage; if the blockage sensitivity coefficient is between 1.2 and 1.5, output according to the maximum membership degree but add a suspected blockage mark.

5. The pilot-operated relief valve fault analysis method according to claim 4, characterized in that, The L1 level high-frequency flutter self-cleaning control method in step 4 is as follows: When the fault level is determined to be mild blockage, the warning indicator light flashes yellow, generating a high-frequency flutter signal. The high-frequency flutter signal is a superposition of a 180 Hz sine wave and Gaussian white noise. The self-cleaning signal is superimposed on the pilot valve's normal control signal. Within 5 seconds after the self-cleaning ends, the signal acquisition, feature extraction, and fault identification steps are re-executed to obtain a new fault level. If the new fault level is normal, the warning is lifted and the self-cleaning is recorded as successful. If it is still mild blockage, the self-cleaning failure counter is incremented. When the number of failures is less than three, the self-cleaning is repeated. When the number of failures reaches three, the fault level is automatically upgraded to moderate blockage, and the corresponding switching control process is initiated.

6. The pilot-operated relief valve fault analysis method according to claim 5, characterized in that, The control method for automatic switching of the backup damping element at level L2 in step 4 is as follows: When the fault level is determined to be moderate blockage, the system pressure is first checked. If the system pressure is higher than the preset switching safety pressure threshold, the pressure is reduced to below the switching safety pressure threshold by bypass unloading or load reduction command. A pulse drive signal with a preset duration is sent to the electromagnet of the damping component automatic switching device. The electromagnet drives the shift fork to rotate, pushing the rotary damping component carrier plate to rotate one fixed position, connecting the spare damping component to the oil circuit, and removing the blocked damping component at the same time. Read the feedback signal from the position sensor installed on the carrier plate. If no position signal is received within the preset waiting time, the switching device is determined to be faulty and the process is directly switched to the severe blockage shutdown procedure. After the switching is completed, wait for a period of stabilization, during which the signal is continuously collected and the blockage sensitivity coefficient is calculated. If the blockage sensitivity coefficient drops below the normal threshold and the fault level drops to normal or mild within the stabilization period, the current state is maintained, the identification information of the original blockage damping component is written into the fault history, and a prompt message is issued; if the blockage sensitivity coefficient is still higher than the normal threshold or the fault level has not dropped below moderate after the stabilization period ends, the process is upgraded to severe blockage shutdown.

7. The pilot-operated relief valve fault analysis method according to claim 6, characterized in that, The control method for L3 level safety shutdown in step 4 is as follows: immediately output a cut-off signal through the main controller to disconnect the main pump power source, requiring the cut-off to be completed in a very short time; simultaneously output an opening signal to the bypass unloading valve to rapidly unload the system pressure in a short time; send a fault latch signal to the equipment main controller, the fault latch signal is a holding type and will not be lost when power is off, and prohibit the equipment from starting; display a serious fault prompt on the human-machine interface, and record the current time, fault level and characteristic data snapshot; After manually troubleshooting, the fault latch signal is cleared by a reset command.

8. The fault analysis method for a pilot-operated relief valve according to claim 7, characterized in that, The method for updating the baseline feature library through self-learning in step 5 is as follows: take several effective feature vectors and their corresponding confirmation labels before the occurrence of the fault as a new sample set; use an incremental single-class support vector machine for learning, with the kernel function being the radial basis function; Calculate the distance metric between each new sample and the existing fault cluster centers of each class. If there exists a class such that the distance is less than a preset multiple of the dispersion of the corresponding class of samples, then merge the new sample into the corresponding class and recalculate the cluster centers and covariance matrix of the corresponding class. If the distances of all classes are greater than or equal to a preset multiple, a new fault subclass is created, a new cluster center is assigned, and the corresponding fault label is associated with it. After a certain number of manual maintenance confirmations are accumulated, the baseline feature library is fully re-clustered to optimize the classification model.

9. The fault analysis method for a pilot-operated relief valve according to claim 8, characterized in that, The cross-correlation time delay in the damping orifice blockage sensitivity coefficient is solved using frequency domain acceleration, including the following sub-steps: Step a: Perform zero-mean processing on the main valve core displacement signal and the pilot valve front chamber pressure signal respectively; Step b: Calculate the cross power spectrum of the two signals, that is, perform Fourier transform on the two signals respectively, and then multiply the Fourier transform of the main valve core displacement with the conjugate of the Fourier transform of the pilot pressure. Step c: Perform an inverse Fourier transform on the cross-power spectrum to obtain the cross-correlation function; Step d: Search for the maximum value of the cross-correlation function within the preset delay time range. The delay time corresponding to the maximum value is the required cross-correlation time delay. In step e, if the calculated delay time exceeds the normal range, the moving average of historical valid values ​​is used instead, and the current calculation is marked as unreliable.

10. A pilot-operated relief valve fault analysis and safety interlock control system, characterized in that, To implement the pilot-operated relief valve fault analysis method of claim 9, the system includes a signal acquisition unit, an active excitation unit, a feature extraction and diagnosis unit, a hierarchical interlocking execution unit, a main controller, and a human-machine interface. The signal acquisition unit includes a first pressure sensor installed at the relief valve inlet, a second pressure sensor installed in the pilot valve front chamber, and a displacement sensor installed on the main valve core, used to synchronize the relief valve inlet pressure signal, the pilot valve front chamber pressure signal, and the main valve core displacement signal at a set sampling frequency. The active excitation unit is used to superimpose the conventional control signal and the excitation signal in real time. The excitation signal includes an active pulse diagnostic excitation signal and a high-frequency chatter self-cleaning signal. The feature extraction and diagnosis unit integrates a wavelet packet decomposition module, a clustering analysis coprocessor, and a non-volatile memory, which is used to store the benchmark feature library. The feature extraction and diagnosis unit is used to extract the feature vector of the real-time signal, perform improved fuzzy C-means clustering calculation, and output the fault type, blockage level and damping orifice blockage sensitivity coefficient. The graded interlocking execution unit includes an early warning indication module, a high-frequency flutter signal generation module, an electromagnetic drive circuit for an automatic damping component switching device, a main pump power source cutoff module, a bypass unloading valve drive module, and a fault latching module. The main controller is an industrial logic controller that communicates bidirectionally with each unit. Its built-in logic program is used to implement the timing control and safety interlocking logic of steps 2 to 4. The human-computer interaction interface is used to display the fault level, sensitivity coefficient, and historical event records, and to receive input of manual confirmation of maintenance tags to trigger the self-learning update process.

Images