A gas transmission and distribution equipment pressure regulator fault early warning and real-time test processing method
By combining multi-dimensional parameter fusion with deep time-series networks, online fault early warning and real-time testing of pressure regulators in gas transmission and distribution systems have been achieved. This solves the technical problems of offline testing affecting gas supply and low early warning accuracy, as well as the inability to adapt to dynamic operating conditions in existing technologies. It enables early fault identification and full life-cycle management, improving the safety and stability of the equipment.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- LESHAN CHUANTIAN GAS EQUIP
- Filing Date
- 2026-02-12
- Publication Date
- 2026-05-12
AI Technical Summary
Existing gas transmission and distribution systems suffer from problems such as offline testing affecting gas supply, low early warning accuracy, inability to adapt to dynamic operating conditions, and lack of full life cycle management, making it difficult to achieve early fault identification and rapid location.
A fault early warning method combining multi-dimensional parameter fusion and deep temporal network is adopted. The operating parameters of the pressure regulator are collected in real time, a dynamic performance feature vector is constructed, and a fault precursor identification is performed through a deep temporal convolutional neural network model to trigger graded early warning. Online real-time testing is carried out without interrupting gas supply, and the remaining lifespan is predicted by combining the full life cycle digital archive.
It enables online non-disruptive testing, improves the accuracy and reliability of fault early warning, adapts to dynamic operating condition changes, reduces operation and maintenance costs, and enhances the safety and stability of equipment.
Smart Images
Figure CN121723327B_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of intelligent manufacturing and industrial safety technology, specifically relating to a method for early warning and real-time testing of faults in pressure regulators of gas transmission and distribution equipment, applicable to intelligent operation and maintenance scenarios for safe production in gas transmission and distribution systems. Background Technology
[0002] Pressure regulators are core equipment in gas transmission and distribution systems. Their function is to regulate the high-pressure gas from upstream to the stable pressure required by downstream users, ensuring a safe and stable gas supply. The operating status of the pressure regulator directly determines the reliability of the gas transmission and distribution system. Once a fault occurs (such as blockage by impurities, diaphragm aging, valve wear, etc.), it may lead to abnormal outlet pressure, gas supply interruption, or even safety accidents such as explosions and poisoning.
[0003] Existing pressure regulator fault detection and early warning technologies have several shortcomings: First, traditional detection methods often employ offline shutdown testing, requiring interruption of normal gas supply, impacting downstream users' production and daily life, and involving long testing cycles, making it difficult to capture early signs of faults in a timely manner. Second, fault identification often relies on single operating parameters (such as outlet pressure), lacking multi-dimensional feature fusion analysis, making it susceptible to interference from operating condition fluctuations, resulting in low early warning accuracy and high false alarm / missed alarm rates. Third, the use of fixed early warning thresholds and static models cannot adapt to dynamic operating conditions such as seasonal changes and gas load fluctuations, leading to a degradation in early warning performance when operating conditions drift. Fourth, there is a lack of a full lifecycle management mechanism, with fault records, test data, and maintenance information being managed in a fragmented manner, making it impossible to predict remaining lifespan and proactively maintain based on historical data. Fifth, fault decision-making lacks interpretability, making it difficult to quickly locate the cause of the fault and formulate a handling plan, thus prolonging the fault handling time.
[0004] Therefore, there is an urgent need for a method for voltage regulator fault early warning and real-time testing that can achieve online non-disruptive testing, accurate fault early warning, adaptability to dynamic operating conditions, and full life cycle management, in order to make up for the deficiencies of existing technologies. Summary of the Invention
[0005] The purpose of this invention is to provide a method for early warning and real-time testing of regulator faults in gas transmission and distribution equipment, which solves the problems of offline testing affecting gas supply, low early warning accuracy, inability to adapt to dynamic operating conditions, and lack of full life cycle management in the prior art, and realizes early identification of regulator faults, online non-disruptive testing, dynamic operating condition adaptation, and intelligent operation and maintenance throughout the entire life cycle.
[0006] To solve the above-mentioned technical problems, the present invention adopts the following solution:
[0007] A method for early warning and real-time testing of faults in pressure regulators of gas transmission and distribution equipment includes the following steps:
[0008] S1. Real-time acquisition of multi-dimensional operating parameters of the pressure regulator: High-precision pressure sensors, flow meters, temperature sensors and vibration acceleration sensors are deployed at the inlet and outlet ends of the pressure regulator to synchronously acquire inlet and outlet pressure, instantaneous flow rate, ambient temperature and shell vibration signal at a rate greater than or equal to the preset sampling frequency, and the multi-dimensional operating parameters are transmitted to the edge computing unit in real time through the industrial bus.
[0009] S2. Construct a dynamic performance feature vector for the pressure regulator: Based on the multi-dimensional operating parameters, calculate the outlet pressure fluctuation rate, differential pressure response lag time, flow-pressure nonlinearity, temperature drift coefficient, and vibration spectrum energy distribution to form a dynamic performance feature vector containing five types of indicators.
[0010] S3. Perform fault precursor identification based on deep temporal network: Input the dynamic performance feature vector into a pre-trained deep temporal convolutional neural network model. The model adopts a multi-layer one-dimensional convolutional layer and a bidirectional long short-term memory unit cascade structure, and outputs the probability distribution of the current state of the voltage regulator as normal, early degradation, mid-term abnormality or serious fault.
[0011] S4. Trigger the graded early warning mechanism: Based on the probability distribution results, if the early degradation probability is greater than or equal to the first preset threshold, a first-level early warning signal is generated and the degradation start time is recorded. If the mid-term abnormal probability is greater than or equal to the second preset threshold, a second-level early warning signal is generated and the online test process is started. If the serious fault probability is greater than or equal to the third preset threshold, a third-level emergency early warning signal is generated and the shut-off valve is linked to perform safety isolation.
[0012] S5. Perform uninterrupted online real-time testing: Without interrupting the gas supply, apply a step pressure excitation with an amplitude of positive and negative predetermined ratio of the rated pressure by adjusting the upstream simulated disturbance device, synchronously record the regulator outlet response curve, calculate its recovery time, overshoot and steady-state error, and compare it with the standard response template for dynamic time warping to generate a performance deviation index.
[0013] S6. Update the equipment health status file: Write the warning level, test results and performance deviation index into the digital file of the voltage regulator's entire life cycle, and predict the remaining effective life based on the historical data sequence using a sliding window regression algorithm. When the predicted remaining life is less than the preset time threshold, a maintenance work order will be automatically pushed.
[0014] The core of this invention lies in its innovative technical solution of accurate fault precursor identification and non-disruptive online real-time testing using multi-dimensional parameter fusion and deep temporal networks. This solution addresses industry pain points such as delayed fault warnings and the need for gas supply interruptions during testing in gas pressure regulators. Existing technologies often rely on single-parameter monitoring, traditional algorithms can only identify overt faults, and fault detection is mostly performed offline with shutdown testing, easily leading to gas supply interruptions and hindering real-time assessment of equipment dynamic performance. This invention simultaneously collects multi-dimensional operating parameters of the pressure regulator (pressure, flow, temperature, vibration) and constructs five types of dynamic performance feature vectors. Combined with a deep temporal network model cascaded with convolutional layers and bidirectional long short-term memory units, it achieves full-stage hierarchical precursor identification from early degradation to severe faults, overcoming the limitation of traditional monitoring that can only detect overt faults. Simultaneously, it designs a non-disruptive step pressure excitation non-disruptive testing method, combined with dynamic time warping comparison to achieve accurate equipment performance assessment. Furthermore, it integrates health records and remaining life prediction to form a "warning-test-maintenance" system. The closed-loop system represents a substantial improvement over existing technologies in terms of the timeliness of fault warning, the practicality of testing, and the intelligence of equipment management. It has outstanding substantive features and significant progress.
[0015] Preferably, in step S2, the outlet pressure fluctuation rate is the ratio of the standard deviation of the outlet pressure to the average pressure over a continuous predetermined time period; the differential pressure response lag time is the time required for the outlet pressure to reach a new steady-state ratio after the inlet pressure changes; the flow-pressure nonlinearity is calculated by fitting the root mean square of the residual between the instantaneous flow rate and the square root of the differential pressure; the temperature drift coefficient is the outlet pressure offset caused by a unit temperature change; and the vibration spectrum energy distribution is obtained by performing a fast Fourier transform on the shell vibration signal and integrating it according to multiple preset frequency bands.
[0016] Preferably, in step S3, the deep temporal convolutional neural network model includes three one-dimensional convolutional layers and two bidirectional long short-term memory units. The first one-dimensional convolutional layer extracts local temporal features, the second one-dimensional convolutional layer further abstracts the features, and the third one-dimensional convolutional layer outputs a feature map which is then input to the bidirectional long short-term memory unit through global average pooling. Finally, it is mapped to the probability distribution of four states through a fully connected layer.
[0017] Preferably, in step S4, the first-level early warning signal is pushed to the mobile terminal of the inspection personnel through the operation and maintenance platform, the second-level early warning signal simultaneously triggers an audible and visual alarm and locks the current operating parameter snapshot, and the third-level emergency early warning signal sends fault location information and suggested handling plan to the regional dispatch center while the linkage shut-off valve performs safety isolation.
[0018] Preferably, in step S5, the simulated disturbance device consists of an electric proportional regulating valve and a gas storage buffer tank, the duration of the step pressure excitation is a predetermined time period, the dynamic time warping comparison adopts Sakoe-Chiba band constraint, the bandwidth is set as a predetermined proportion of the sequence length, and the performance deviation index is the ratio of the cumulative distance of the warped path to the length of the standard template.
[0019] Preferably, in step S6, the full lifecycle digital archive is stored in a cloud database and includes the installation date, cumulative runtime, all warning records, test reports and maintenance logs. The sliding window regression algorithm uses ridge regression, the window length is a predetermined time period, and the remaining effective lifespan prediction update cycle is once every predetermined time interval.
[0020] Preferably, it also includes establishing a knowledge graph of regulator failure modes. The knowledge graph is constructed based on a historical failure case library and includes a causal relationship network of five typical failures: impurity blockage, diaphragm aging, spring fatigue, valve wear, and controller failure. Each failure is associated with an abnormal combination pattern of at least three characteristic indicators, which is used to assist the decision interpretation and false alarm filtering of the deep time series network.
[0021] Preferably, it also includes an adaptive threshold calibration mechanism, which dynamically adjusts the early warning probability threshold according to seasonal changes and gas load characteristics. When the daily average flow rate changes within a predetermined number of consecutive days exceed a preset proportion, the model fine-tuning process is automatically initiated, and online transfer learning is performed on the network output layer using fault-free data from the most recent predetermined time period.
[0022] The beneficial effects of this invention are as follows:
[0023] 1. Enables uninterrupted online testing: Dynamic performance testing of the pressure regulator can be completed without interrupting the gas supply, avoiding the impact of offline testing on downstream gas supply, while capturing the equipment operating status in real time and promptly detecting early signs of failure;
[0024] 2. High accuracy and reliability of early warning: By using multi-dimensional feature fusion, deep temporal network modeling and fault knowledge graph assistance, combined with adaptive threshold calibration, the false alarm and false negative rates are effectively reduced, adapting to dynamic operating condition changes and improving the accuracy of fault identification and decision interpretability.
[0025] 3. Intelligent management and control throughout the entire lifecycle: Construct a digital archive for the entire lifecycle, predict the remaining effective lifespan based on historical data, promote the transformation of operation and maintenance mode from "post-failure repair" to "predictive maintenance", and reduce operation and maintenance costs;
[0026] 4. Lightweight, efficient, and highly practical: Model fine-tuning is only applied to the output layer, resulting in low computational load and short processing time. It is suitable for edge computing scenarios, and the hierarchical early warning mechanism and linkage handling function ensure rapid fault response, thereby improving the safety and stability of the gas transmission and distribution system. Attached Figure Description
[0027] Figure 1 This is a flowchart of a method for early warning and real-time testing of a pressure regulator fault in a gas transmission and distribution equipment according to the present invention. Detailed Implementation
[0028] The present invention will be further described in detail below with reference to the embodiments and accompanying drawings, but the embodiments of the present invention are not limited thereto.
[0029] like Figure 1 As shown, Figure 1 This is a flowchart illustrating a method for early warning and real-time testing of a pressure regulator fault in gas transmission and distribution equipment, according to the present invention. The method includes the following steps:
[0030] S1. Real-time acquisition of multi-dimensional operating parameters of the pressure regulator: High-precision pressure sensors, flow meters, temperature sensors and vibration acceleration sensors are deployed at the inlet and outlet ends of the pressure regulator to synchronously acquire inlet and outlet pressure, instantaneous flow rate, ambient temperature and shell vibration signal at a rate greater than or equal to the preset sampling frequency, and the multi-dimensional operating parameters are transmitted to the edge computing unit in real time through the industrial bus.
[0031] S2. Construct a dynamic performance feature vector for the pressure regulator: Based on the multi-dimensional operating parameters, calculate the outlet pressure fluctuation rate, differential pressure response lag time, flow-pressure nonlinearity, temperature drift coefficient, and vibration spectrum energy distribution to form a dynamic performance feature vector containing five types of indicators.
[0032] S3. Perform fault precursor identification based on deep temporal network: Input the dynamic performance feature vector into a pre-trained deep temporal convolutional neural network model. The model adopts a multi-layer one-dimensional convolutional layer and a bidirectional long short-term memory unit cascade structure, and outputs the probability distribution of the current state of the voltage regulator as normal, early degradation, mid-term abnormality or serious fault.
[0033] S4. Trigger the graded early warning mechanism: Based on the probability distribution results, if the early degradation probability is greater than or equal to the first preset threshold, a first-level early warning signal is generated and the degradation start time is recorded. If the mid-term abnormal probability is greater than or equal to the second preset threshold, a second-level early warning signal is generated and the online test process is started. If the serious fault probability is greater than or equal to the third preset threshold, a third-level emergency early warning signal is generated and the shut-off valve is linked to perform safety isolation.
[0034] S5. Perform uninterrupted online real-time testing: Without interrupting the gas supply, apply a step pressure excitation with an amplitude of positive and negative predetermined ratio of the rated pressure by adjusting the upstream simulated disturbance device, synchronously record the regulator outlet response curve, calculate its recovery time, overshoot and steady-state error, and compare it with the standard response template for dynamic time warping to generate a performance deviation index.
[0035] S6. Update the equipment health status file: Write the warning level, test results and performance deviation index into the digital file of the voltage regulator's entire life cycle, and predict the remaining effective life based on the historical data sequence using a sliding window regression algorithm. When the predicted remaining life is less than the preset time threshold, a maintenance work order will be automatically pushed.
[0036] Preferably, in step S2, the outlet pressure fluctuation rate is the ratio of the standard deviation of the outlet pressure to the average pressure over a continuous predetermined time period; the differential pressure response lag time is the time required for the outlet pressure to reach a new steady-state ratio after the inlet pressure changes; the flow-pressure nonlinearity is calculated by fitting the root mean square of the residual between the instantaneous flow rate and the square root of the differential pressure; the temperature drift coefficient is the outlet pressure offset caused by a unit temperature change; and the vibration spectrum energy distribution is obtained by performing a fast Fourier transform on the shell vibration signal and integrating it according to multiple preset frequency bands.
[0037] Preferably, in step S3, the deep temporal convolutional neural network model includes three one-dimensional convolutional layers and two bidirectional long short-term memory units. The first one-dimensional convolutional layer extracts local temporal features, the second one-dimensional convolutional layer further abstracts the features, and the third one-dimensional convolutional layer outputs a feature map which is then input to the bidirectional long short-term memory unit through global average pooling. Finally, it is mapped to the probability distribution of four states through a fully connected layer.
[0038] Preferably, in step S4, the first-level early warning signal is pushed to the mobile terminal of the inspection personnel through the operation and maintenance platform, the second-level early warning signal simultaneously triggers an audible and visual alarm and locks the current operating parameter snapshot, and the third-level emergency early warning signal sends fault location information and suggested handling plan to the regional dispatch center while the linkage shut-off valve performs safety isolation.
[0039] Preferably, in step S5, the simulated disturbance device consists of an electric proportional regulating valve and a gas storage buffer tank, the duration of the step pressure excitation is a predetermined time period, the dynamic time warping comparison adopts Sakoe-Chiba band constraint, the bandwidth is set as a predetermined proportion of the sequence length, and the performance deviation index is the ratio of the cumulative distance of the warped path to the length of the standard template.
[0040] Preferably, in step S6, the full lifecycle digital archive is stored in a cloud database and includes the installation date, cumulative runtime, all warning records, test reports and maintenance logs. The sliding window regression algorithm uses ridge regression, the window length is a predetermined time period, and the remaining effective lifespan prediction update cycle is once every predetermined time interval.
[0041] Preferably, it also includes establishing a knowledge graph of regulator failure modes. The knowledge graph is constructed based on a historical failure case library and includes a causal relationship network of five typical failures: impurity blockage, diaphragm aging, spring fatigue, valve wear, and controller failure. Each failure is associated with an abnormal combination pattern of at least three characteristic indicators, which is used to assist the decision interpretation and false alarm filtering of the deep time series network.
[0042] Preferably, it also includes an adaptive threshold calibration mechanism, which dynamically adjusts the early warning probability threshold according to seasonal changes and gas load characteristics. When the daily average flow rate changes within a predetermined number of consecutive days exceed a preset proportion, the model fine-tuning process is automatically initiated, and online transfer learning is performed on the network output layer using fault-free data from the most recent predetermined time period.
[0043] The present invention will be further described in detail below with reference to specific embodiments. This embodiment takes a building-type pressure regulator (rated inlet pressure 0.4MPa, rated outlet pressure 2kPa) commonly used in urban gas transmission and distribution systems as an example.
[0044] Step S1: Multidimensional Operation Parameter Acquisition
[0045] A pressure sensor with an accuracy of ±0.01MPa is deployed at the inlet of the pressure regulator, and a pressure sensor with an accuracy of ±0.001MPa and a flow meter with a range of 0-10m³ / h are deployed at the outlet. A vibration acceleration sensor (range 0-50g, frequency response 0-500Hz) and an ambient temperature sensor (range -40℃-85℃, accuracy ±0.1℃) are deployed on the housing surface. The sampling frequency is set to 200Hz to simultaneously collect inlet pressure, outlet pressure, instantaneous flow rate, ambient temperature, and housing vibration signals, which are transmitted to the edge computing unit (using an Intel Core i5 processor and 8GB of memory) via a ModbusTCP industrial bus.
[0046] Step S2: Construction of Dynamic Performance Feature Vectors
[0047] A continuous predetermined time period of 10 minutes was set, and the outlet pressure fluctuation rate was calculated. After the inlet pressure changed, the time required for the outlet pressure to reach 95% of the new steady state was taken as the differential pressure response lag time. The relationship curve between instantaneous flow rate and the square root of differential pressure was fitted, and the root mean square of the residual was calculated as the flow-pressure nonlinearity. The temperature drift coefficient was calculated in units of 1℃. The vibration signal was subjected to FFT transformation and divided into five frequency bands: 5Hz-20Hz, 20Hz-40Hz, 40Hz-60Hz, 60Hz-80Hz, and 80Hz-100Hz. The vibration spectrum energy distribution was obtained by integration, forming a 5-dimensional dynamic performance feature vector.
[0048] Step S3: Running the Fault Precursor Identification Model
[0049] The parameters of the three one-dimensional convolutional layers in the deep temporal convolutional neural network model are as follows: Layer 1 (3×1 kernels, 32 units, stride 1, ReLU activation function), Layer 2 (5×1 kernels, 64 units, stride 1, ReLU activation function), and Layer 3 (3×1 kernels, 64 units, stride 1, ReLU activation function); the output dimension after global average pooling is [batch size, 64]; each of the two BiLSTM layers has 128 hidden units and a dropout coefficient of 0.2; the fully connected layer has an output dimension of 4 and uses the Softmax activation function. The model is pre-trained using 15,000 samples (containing four states: normal, early degradation, mid-term abnormality, and severe failure, with 3,750 samples per state). After training, the samples are deployed to the edge computing unit, and the feature vector constructed in step S2 is input into the model to output the probability distribution of the four states.
[0050] Step S4: Triggering of Tiered Early Warning
[0051] The system sets a first preset threshold of 70%, a second preset threshold of 80%, and a third preset threshold of 90%. If the model outputs an early degradation probability of 75%, a Level 1 warning signal is generated and pushed to the inspection personnel's mobile app via the operation and maintenance platform, recording the degradation start time. If the mid-term anomaly probability is 82%, a Level 2 warning signal is generated, triggering an on-site audible and visual alarm and locking a snapshot of the current operating parameters. If the probability of a serious fault is 91%, a Level 3 emergency warning signal is generated, triggering the closure of the inlet shut-off valve and sending fault location (building number, regulator location), real-time operating data, and handling suggestions (replacing the regulator diaphragm) to the regional dispatch center.
[0052] Step S5: Uninterrupted online real-time testing
[0053] The simulated disturbance device consists of a DN25 electric proportional control valve (adjustment accuracy ±0.5%FS) and a 0.5m³ gas storage buffer tank. A step pressure excitation with an amplitude of ±10% of the rated inlet pressure (i.e., 0.36MPa and 0.44MPa) is applied for 20s. The outlet pressure response curve is recorded simultaneously, and the recovery time is calculated to be 2.5s, the overshoot is 5%, and the steady-state error is ±0.05kPa. The standard response template uses the test curve of the new equipment of this type of pressure regulator. The DTW comparison uses Sakoe-Chiba band constraint, and the bandwidth is set to 8% of the sequence length (4000 time steps) (320 time steps). The calculated regularized path cumulative distance is 12.8, the standard template length is 4000, and the performance deviation index is 12.8 / 4000=0.0032.
[0054] Step S6: Health Record Update and Life Prediction
[0055] The Level 2 warning signal, recovery time of 2.5s, overshoot of 5%, steady-state error of ±0.05kPa, and performance deviation index of 0.0032 are written into the cloud-based full lifecycle digital archive. Ridge regression is used for sliding window regression with a window length of 120 days and a regularization parameter of λ=0.1. The remaining effective lifespan prediction update cycle is 10 days. If the predicted remaining lifespan is 60 days (greater than the preset threshold of 30 days), no maintenance work order will be pushed, and monitoring will continue. If the predicted remaining lifespan is 25 days, a maintenance work order will be automatically pushed to the operation and maintenance platform.
[0056] Step S7: Knowledge Graph-Assisted Decision Making
[0057] In the fault mode knowledge graph, the abnormal feature combination associated with diaphragm aging is: increased outlet pressure fluctuation rate (>0.02), prolonged differential pressure response hysteresis time (>3s), and higher energy in the 20Hz-40Hz frequency band of the vibration spectrum. If the model identifies it as an intermediate-term anomaly, the fault type is determined to be diaphragm aging by comparing with the knowledge graph, thus filtering out false alarms caused by temperature fluctuations.
[0058] Step S8: Adaptive calibration and model fine-tuning
[0059] During the peak winter gas consumption period, the average daily flow rate increases by 25% compared to the off-season. The adaptive mechanism adjusts the first preset threshold to 75% and the second preset threshold to 85%. If the average daily flow rate changes by 28% for 5 consecutive days, the model fine-tuning process is automatically initiated. Using the most recent 45 days of fault-free data (1000 sets), only the weights of the fully connected layer are updated. The fine-tuning takes 20 minutes. After completion, the model adapts to high-load conditions, and the early warning accuracy is improved by 12%.
[0060] This invention also includes a system for implementing fault early warning and real-time testing and processing methods for pressure regulators in gas transmission and distribution equipment. The system includes a pressure regulator, an inlet pressure sensor, an outlet pressure sensor, a flow meter, a temperature sensor, a vibration acceleration sensor, an edge computing unit, an industrial bus, a simulated disturbance device, a shut-off valve, a deep temporal convolutional neural network model, a full lifecycle digital archive, an operation and maintenance platform, and a regional dispatch center. All the above components form a complete closed loop of monitoring, analysis, early warning, and execution through physical connections or communication links.
[0061] In specific deployments, the pressure regulator is a key pressure stabilizing device in the gas transmission and distribution network. Its inlet is connected to the upstream gas supply pipeline, and its outlet is connected to the downstream user network. An inlet pressure sensor is installed near the inlet flange of the pressure regulator to collect upstream gas supply pressure in real time; an outlet pressure sensor is installed near the outlet flange of the pressure regulator to monitor the downstream pressure after regulation. A flow meter, using an ultrasonic or turbine structure, is connected in series on the outlet pipe of the pressure regulator to measure instantaneous volumetric flow rate. A temperature sensor is fixed to the outer surface of the pressure regulator housing or embedded in the internal cavity wall to sense changes in ambient or medium temperature. A vibration acceleration sensor is bolted to the top or side wall of the pressure regulator housing to collect mechanical vibration signals generated during equipment operation. All five types of sensors are connected to an industrial bus via shielded cables. The industrial bus uses Modbus TCP or PROFINET protocols to transmit synchronously collected multi-dimensional operating parameters to the edge computing unit at a sampling frequency of no less than 100Hz.
[0062] The present invention will be further described in detail below with reference to specific embodiments. This embodiment takes a building-type pressure regulator (rated inlet pressure 0.4MPa, rated outlet pressure 2kPa) commonly used in urban gas transmission and distribution systems as an example. Ten pressure regulators of the same model in the same area were selected as experimental objects. Five of them were monitored and maintained using the method of the present invention (experimental group), and the other five were tested using the traditional offline periodic testing method (control group, shut down for testing once every 3 months). The experimental period was 1 year. The experimental data and results are described below:
[0063] I. Experimental Parameter Setting and Data Acquisition
[0064] (a) Basic parameter configuration
[0065] Sensor parameters: Inlet pressure sensor accuracy ±0.01MPa, outlet pressure sensor accuracy ±0.001MPa, flow metering range 0-10m 3 / h, vibration acceleration sensor range 0-50g, frequency response 0-500Hz, ambient temperature sensor range -40℃-85℃, accuracy ±0.1℃.
[0066] Sampling and transmission parameters: Sampling frequency 200Hz, transmitted to edge computing unit (Intel Core i5 processor, 8GB memory) via Modbus TCP industrial bus.
[0067] Feature calculation parameters: a continuous predetermined time period of 10 minutes (used for calculating the outlet pressure fluctuation rate), and the outlet pressure reaching 95% of the new steady state is used as the criterion for judging the pressure difference response lag time. After FFT transformation, the vibration signal is divided into five frequency bands: 5Hz-20Hz, 20Hz-40Hz, 40Hz-60Hz, 60Hz-80Hz, and 80Hz-100Hz.
[0068] Model parameters: The deep temporal convolutional neural network consists of three one-dimensional convolutional layers (parameters are 3×1 kernel, number of layers 32, stride 1, activation function ReLU; 5×1 kernel, number of layers 64, stride 1, activation function ReLU; 3×1 kernel, number of layers 64, stride 1, activation function ReLU), two BiLSTM layers (each with 128 hidden units and dropout coefficient 0.2), and a fully connected layer with an output dimension of 4 (activation function Softmax). 15,000 pre-trained samples were used (3,750 samples for each of the four states).
[0069] Warning thresholds: First preset threshold 70% (early degradation), second preset threshold 80% (mid-term anomaly), third preset threshold 90% (serious failure).
[0070] Undisturbed test parameters: The simulated disturbance device consists of a DN25 electric proportional control valve (adjustment accuracy ±0.5% FS) and a 0.5m... 3 The system consists of a gas storage buffer tank, with a step pressure excitation amplitude of ±10% of the rated inlet pressure (0.36MPa and 0.44MPa) and a duration of 20s. The DTW comparison uses Sakoe-Chiba band constraints with a bandwidth of 8% of the sequence length (4000 time steps) (320 time steps).
[0071] Lifetime prediction parameters: Ridge regression is used for sliding window regression (regularization parameter λ=0.1), window length is 120 days, remaining effective lifetime prediction update cycle is 10 days, and maintenance work order push threshold is 30 days.
[0072] Adaptive calibration parameters: When the daily average flow rate changes by more than the preset proportion (20%) for 5 consecutive days, the model fine-tuning is initiated. The weights of the fully connected layer are updated using the most recent 45 days of fault-free data (1000 sets), and the fine-tuning time is 20 minutes.
[0073] (II) Experimental Data Collection Results
[0074] Experimental group data: Approximately 8.76 × 10⁻⁶ valid operational data were collected within one year. 7 3.5 × 10⁻⁶ lines are generated to form a dynamic performance feature vector. 4 The system triggered Level 1 warnings 12 times, Level 2 warnings 8 times, and Level 3 warnings 2 times. A total of 18 uninterrupted online tests were performed, with outlet pressure fluctuations controlled within ±0.08 kPa during the tests. No gas supply interruptions occurred, and the performance deviation index ranged from 0.0021 to 0.0058.
[0075] Control group data: Four offline shutdown tests were performed within one year, with each test interrupting the gas supply for about 40 minutes. Three mid-term anomalies and two serious faults were found during the tests, but no early degradation state was captured. The test data only includes static performance parameters under shutdown conditions and does not include dynamic characteristic data during continuous operation.
[0076] II. Fault Identification and Early Warning Effectiveness Data
[0077] (a) Fault Types and Identification Status
[0078] During the experimental period, the actual types and numbers of failures that occurred in the 10 voltage regulators were as follows: 3 cases of impurity blockage, 4 cases of diaphragm aging, 2 cases of spring fatigue, 1 case of valve port wear, and 1 case of pilot malfunction, totaling 11 failures (including the early degradation stage).
[0079] Table 1: Comparison of identification results between the experimental and control groups for various fault types of voltage regulators
[0080]
[0081] (II) Data on the accuracy and timeliness of early warning
[0082] Early warning accuracy: In the experimental group, 11 out of 12 Level 1 early warnings were confirmed as true early degradation (false alarm rate 8.3%); 8 Level 2 early warnings were confirmed as true mid-term anomalies (false alarm rate 0%); and 2 Level 3 early warnings were confirmed as true serious faults (false alarm rate 0%). There were no missed warnings. The control group had no false alarms, but the missed warning rate was 54.5% (6 faults were not identified).
[0083] Early warning timeliness: The experimental group had an average early warning time of 23 days for early degradation faults and an average early warning time of 8 days for mid-term abnormal faults; the control group could only detect faults during offline testing and had no early warning capability. By the time serious faults were detected, they had already caused two minor pressure abnormal fluctuations (outlet pressure overshoot reached 15%).
[0084] III. Testing and Maintenance Results Data
[0085] (a) Comparison of test impacts
[0086] Table 2: Comparison of test indicators between the experimental and control groups for voltage regulators
[0087]
[0088] (ii) Operation and maintenance costs and equipment lifespan data
[0089] Maintenance Costs: The experimental group, due to early warnings and predictive maintenance, replaced 3 diaphragms, cleaned impurities 3 times, and replaced 1 spring within one year. The total maintenance material cost was 8600 yuan, and the labor cost (including inspection and maintenance) was 12000 yuan, for a total maintenance cost of 20600 yuan. The control group, due to repairs after malfunctions, replaced 2 diaphragms, cleaned impurities once, replaced 1 spring, replaced 1 valve port, and replaced 1 controller. The total maintenance material cost was 11200 yuan, and the labor cost (including emergency repairs and downtime testing) was 25000 yuan, for a total maintenance cost of 36200 yuan. The experimental group's maintenance costs were 43.1% lower than the control group.
[0090] Equipment lifespan: At the end of the experiment, all 5 voltage regulators in the experimental group were operating normally, and the predicted remaining effective lifespan of each group was more than 45 days. One voltage regulator in the control group was scrapped due to a serious malfunction that was not dealt with in time, resulting in damage to the casing. The predicted remaining effective lifespan of the other 4 voltage regulators was an average of 22 days, and the average equipment lifespan was about 15% shorter than that of the experimental group.
[0091] (III) Adaptive calibration effect data
[0092] During the peak gas consumption period in winter (November-February), the average daily flow rate in the experimental group increased by 28%-35% compared to the off-season. The adaptive mechanism automatically adjusted the early warning thresholds (75% for Level 1 and 85% for Level 2) and initiated model fine-tuning three times. After fine-tuning, the fault identification accuracy improved from 91.7% in the off-season to 95.8%, and the false alarm rate decreased from 8.3% in the off-season to 3.2%. In contrast, the control group, lacking a dynamic adaptation mechanism, failed to identify any early degradation faults during the winter, resulting in a 20% increase in the missed alarm rate compared to the off-season.
[0093] IV. Effect Description
[0094] Comprehensiveness and accuracy of fault identification: This invention achieves full-stage fault identification from early degradation to severe faults through multi-dimensional parameter fusion and deep temporal network modeling, with a total fault identification rate of 100% and a false alarm rate of only 8.3%. Compared with traditional offline testing (identification rate of 45.5% and false alarm rate of 54.5%), it significantly improves the comprehensiveness and accuracy of fault identification, especially solving the industry pain point of difficulty in capturing early degradation faults.
[0095] Practicality of Disturbance Testing: The disturbance-free online testing solution can quickly complete the dynamic performance evaluation of equipment without interrupting the gas supply (0 minutes of gas supply interruption per year). The test time is only 20 seconds, which is much shorter than the 40 minutes of traditional offline testing. Moreover, the pressure fluctuation during the test is controlled within ±0.08 kPa, so that the user does not feel it. This completely solves the problem of traditional offline testing affecting the downstream gas supply.
[0096] Dynamic operating condition adaptability: The adaptive threshold calibration and model fine-tuning mechanism can dynamically adjust the early warning strategy and model parameters according to seasonal changes and gas load fluctuations. During the peak winter period, the fault identification accuracy is improved by 4.1%, and the false alarm rate is reduced by 5.1%, effectively avoiding the degradation of early warning performance caused by operating condition drift, and adapting to the dynamic operating characteristics of the gas transmission and distribution system.
[0097] Operational efficiency and system security: Through predictive maintenance, the experimental group reduced operation and maintenance costs by 43.1% compared to the control group, extended the average lifespan of equipment by about 15%, and did not experience any safety incidents caused by serious failures. The control group, which relied on repairs after failures, not only incurred high operation and maintenance costs but also experienced two abnormal pressure fluctuations and the scrapping of one piece of equipment. This verifies the significant advantages of the "early warning-testing-maintenance" closed-loop mechanism of this invention in reducing operation and maintenance costs and ensuring the safe and stable operation of the system.
[0098] In summary, through technological innovations such as multi-dimensional parameter acquisition, deep time-series network fault identification, non-disruptive online testing, full life-cycle management and adaptive calibration, this invention has achieved significant results in terms of timely fault warning, practical testing, adaptability to operating conditions, and intelligent operation and maintenance, providing reliable technical support for the safe and efficient operation and maintenance of pressure regulators in gas transmission and distribution systems.
[0099] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Based on the technical essence of the present invention, any simple modifications, equivalent substitutions, and improvements made to the above embodiments within the spirit and principles of the present invention shall still fall within the protection scope of the present invention.
Claims
1. A method for early warning and real-time testing and processing of faults in pressure regulators of gas transmission and distribution equipment, characterized in that, Includes the following steps: S1. Real-time acquisition of multi-dimensional operating parameters of the pressure regulator: High-precision pressure sensors, flow meters, temperature sensors and vibration acceleration sensors are deployed at the inlet and outlet ends of the pressure regulator to synchronously acquire inlet and outlet pressure, instantaneous flow rate, ambient temperature and shell vibration signal at a rate greater than or equal to the preset sampling frequency, and the multi-dimensional operating parameters are transmitted to the edge computing unit in real time through the industrial bus. S2. Constructing the dynamic performance feature vector of the pressure regulator: Based on the multi-dimensional operating parameters, calculate the outlet pressure fluctuation rate, differential pressure response lag time, flow-pressure nonlinearity, temperature drift coefficient, and vibration spectrum energy distribution to form a dynamic performance feature vector containing five types of indicators; the outlet pressure fluctuation rate is the ratio of the standard deviation of the outlet pressure to the average pressure within a continuous predetermined time period; the differential pressure response lag time is the time required for the outlet pressure to reach 95% of the new steady state after the inlet pressure changes; the flow-pressure nonlinearity is calculated by fitting the root mean square of the residual between the instantaneous flow rate and the square root of the differential pressure; the temperature drift coefficient is the outlet pressure offset caused by a unit temperature change; and the vibration spectrum energy distribution is obtained by performing a fast Fourier transform on the shell vibration signal and integrating it according to multiple preset frequency bands. S3. Perform fault precursor identification based on deep temporal networks: Input the dynamic performance feature vector into a pre-trained deep temporal convolutional neural network model. This model adopts a cascaded structure of multiple one-dimensional convolutional layers and bidirectional long short-term memory units, and outputs the probability distribution of the current state of the voltage regulator as normal, early degradation, mid-term abnormality, or severe fault. The deep temporal convolutional neural network model contains three one-dimensional convolutional layers and two bidirectional long short-term memory units. The first one-dimensional convolutional layer extracts local temporal features, the second one-dimensional convolutional layer further abstracts the features, and the third one-dimensional convolutional layer outputs a feature map which is then input to the bidirectional long short-term memory unit through global average pooling. Finally, it is mapped to the probability distribution of the four states through a fully connected layer. S4. Trigger the graded early warning mechanism: Based on the probability distribution results, if the early degradation probability is greater than or equal to the first preset threshold, a first-level early warning signal is generated and the degradation start time is recorded. If the mid-term abnormal probability is greater than or equal to the second preset threshold, a second-level early warning signal is generated and the online test process is started. If the serious fault probability is greater than or equal to the third preset threshold, a third-level emergency early warning signal is generated and the shut-off valve is linked to perform safety isolation. S5. Perform uninterrupted online real-time testing: Without interrupting the gas supply, apply a step pressure excitation with an amplitude of positive and negative predetermined ratio of the rated pressure by adjusting the upstream simulated disturbance device, synchronously record the regulator outlet response curve, calculate its recovery time, overshoot and steady-state error, and compare it with the standard response template for dynamic time warping to generate a performance deviation index. S6. Update the equipment health status file: Write the warning level, test results and performance deviation index into the digital file of the voltage regulator's entire life cycle, and predict the remaining effective life based on the historical data sequence using a sliding window regression algorithm. When the predicted remaining life is less than the preset time threshold, a maintenance work order will be automatically pushed.
2. The method for fault early warning and real-time testing and processing of pressure regulators in gas transmission and distribution equipment according to claim 1, characterized in that, In step S4, the first-level early warning signal is pushed to the mobile terminal of the inspection personnel through the operation and maintenance platform. The second-level early warning signal simultaneously triggers an audible and visual alarm and locks the current operating parameter snapshot. The third-level emergency early warning signal sends fault location information and suggested handling plan to the regional dispatch center while the linkage shut-off valve performs safety isolation.
3. The method for early warning and real-time testing of pressure regulator faults in gas transmission and distribution equipment according to claim 1, characterized in that, In step S5, the simulated disturbance device consists of an electric proportional regulating valve and a gas storage buffer tank. The duration of the step pressure excitation is a predetermined time period. The dynamic time warping comparison adopts Sakoe-Chiba band constraint, and the bandwidth is set as a predetermined proportion of the sequence length. The performance deviation index is the ratio of the cumulative distance of the warped path to the length of the standard template.
4. The method for early warning and real-time testing of pressure regulator faults in gas transmission and distribution equipment according to claim 1, characterized in that, In step S6, the full lifecycle digital archive is stored in a cloud database, including the installation date, cumulative runtime, all warning records, test reports and maintenance logs. The sliding window regression algorithm uses ridge regression, the window length is a predetermined time period, and the remaining effective lifespan prediction update cycle is once every predetermined time interval.
5. The method for fault early warning and real-time testing and processing of pressure regulators in gas transmission and distribution equipment according to claim 1, characterized in that, It also includes establishing a knowledge graph of regulator failure modes. The knowledge graph is built based on a historical failure case library and contains a causal relationship network of five typical failures: impurity blockage, diaphragm aging, spring fatigue, valve wear, and pilot malfunction. Each failure is associated with an abnormal combination pattern of at least three characteristic indicators, which is used to assist the decision interpretation and false alarm filtering of deep time series networks.
6. The method for early warning and real-time testing of pressure regulator faults in gas transmission and distribution equipment according to claim 1, characterized in that, It also includes an adaptive threshold calibration mechanism that dynamically adjusts the early warning probability threshold based on seasonal changes and gas load characteristics. When the daily average flow rate changes within a predetermined number of consecutive days exceed a preset ratio, the model fine-tuning process is automatically initiated, and online transfer learning is performed on the network output layer using fault-free data from the most recent predetermined time period.