An online multimodal intelligent leak monitoring and alarm device for turbine rubber sealing rings
By combining multimodal sensors and edge computing with blockchain technology, the issues of real-time performance, accuracy, and data security in monitoring turbine rubber sealing rings have been resolved. This has enabled highly sensitive leak detection and reliable fault tracing, while reducing maintenance costs.
Patent Information
- Application Number
- CN202610978385.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-07-02
- Publication Date
- 2026-11-10
AI Technical Summary
Existing technologies cannot achieve real-time and accurate monitoring of turbine rubber sealing rings. Sensor signals are singular and easily interfered with, leading to frequent false alarms and missed alarms. Furthermore, data is easily tampered with and difficult to trace, resulting in high maintenance costs.
Employing an eight-channel differential pressure sensor, six circumferential ultrasonic transducers, and a polyaniline-modified carbon nanotube electrochemical sensing array, combined with edge computing and blockchain technology, multimodal information fusion monitoring is achieved. The stability of the sensor and data security are ensured through self-calibration and energy harvesting modules.
It achieves highly sensitive, low-delay differential detection of leakage and material aging in turbine rubber sealing rings, reduces false alarms and missed alarms, ensures data immutability and reduces maintenance costs, and provides reliable fault tracing capabilities.
Smart Images

Figure CN123065684A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of steam turbine sealing technology, and specifically relates to an online multimodal intelligent leakage monitoring and alarm device for steam turbine rubber sealing rings. Background Technology
[0002] As a crucial sealing element between the flow-through and atmospheric sides of a steam turbine, the rubber sealing ring typically operates under extreme conditions, including high temperatures, low negative pressure, highly corrosive steam, and high-frequency vibration. During long-term operation, the sealing ring material is prone to rubber aging, hardening, cracking, and localized wear, leading to changes in the pressure difference between the inside and outside of the sealing cavity, steam leakage, and a decrease in vacuum. In extreme cases, this can result in abnormal shutdowns. Traditional manual inspections and single-sensor monitoring methods, such as single-channel differential pressure, temperature, or ultrasonic probes, have the following shortcomings: 1. Poor response time: Manual inspection has a long cycle and cannot achieve real-time monitoring, while single-sensor online monitoring has a single sensing signal and cannot accurately distinguish different fault modes such as leakage and material deterioration.
[0003] 2. Insufficient sensitivity and reliability: Single differential pressure or ultrasonic monitoring has limited sensitivity to microcracks of 10µm level, and the signal is easily affected by noise and temperature drift, resulting in frequent false alarms and missed alarms.
[0004] 3. Calibration and maintenance are difficult: Under high temperature and high pressure environments, the sensor experiences rapid zero-point drift and sensitivity decay, and traditional periodic manual calibration is costly and risky.
[0005] 4. Lack of data security and traceability: Monitoring data is easily tampered with or lost, making it difficult to meet the power plant's requirements for safety audits and fault tracing.
[0006] In view of the above problems, there is an urgent need for an online intelligent leak monitoring and alarm device that comprehensively utilizes multimodal information such as differential pressure, ultrasonic and electrochemical sensing, and combines edge intelligent computing and secure evidence storage technology to achieve high sensitivity, low latency, multi-level, self-calibrating and traceable characteristics. Summary of the Invention
[0007] To address the shortcomings of existing technologies, this invention provides an online multimodal intelligent leak monitoring and alarm device for turbine rubber sealing rings. By combining three types of sensors—an eight-channel differential pressure sensor, six circumferential ultrasonic transducers, and a polyaniline-modified carbon nanotube electrochemical sensor array—it comprehensively captures multi-source characteristic information of sealing ring leaks, enabling the distinguishable detection of leaks and material aging, thereby reducing false alarms and missed alarms.
[0008] To achieve the objective of this invention, the technical solution adopted is as follows: An online multimodal intelligent leak monitoring and alarm device for turbine rubber sealing rings includes: an eight-channel differential pressure sensor, six circumferential ultrasonic transducers, a polyaniline-modified carbon nanotube electrochemical sensing array, an edge computing unit, a blockchain security storage module, a local alarm execution unit, a multi-mode communication interface, a self-calibration module, and an energy harvesting module.
[0009] An eight-channel differential pressure sensor is arranged at equal intervals along the inner and outer cavities of the sealing ring to collect pressure difference signals, temperature and humidity signals in real time. The pressure difference signal includes the average pressure difference and the amplitude characteristics of the pressure difference fluctuation. The preferred eight-channel differential pressure sensor is the P2X700, with a range of 0 kPa to 100 kPa, an accuracy of 0.1%FS, and a temperature resistance of -40℃ to 120℃.
[0010] Six circumferential ultrasonic transducers are uniformly installed along the circumference to collect ultrasonic signals generated by leaks. These transducers transmit 1 µs pulses in a time-division multiplexing manner and receive changes in the standing wave envelope. The ultrasonic signals generated by leaks are collected by extracting the standing wave peak amplitude change ΔA and phase difference spectrum characteristics. The ultrasonic signals include the standing wave peak amplitude and phase difference spectrum characteristics. When microcracks, wear, or steam leakage occur in the sealing area, the ultrasonic propagation path and acoustic impedance change, leading to attenuation of the standing wave peak amplitude and phase shift. When the standing wave peak amplitude ΔA > 3%, combined with phase difference spectrum analysis, microcrack detection at the 10 µm level is achieved, with an algorithm response delay ≤ 20 ms. The preferred center frequency for the six circumferential ultrasonic transducers is 1 MHz, with a bandwidth of ±100 kHz.
[0011] A polyaniline-modified carbon nanotube electrochemical sensing array is configured in the steam sampling branch to detect electrochemical signals online; the electrochemical signals include H2O2 and Fe²⁺. + Concentration change rate characteristics; Electrode area of polyaniline-modified carbon nanotube electrochemical sensing array is 1 cm².
[0012] The edge computing unit, based on the STM32H743 microprocessor and NPU coprocessor, incorporates a multimodal fusion neural network and an adaptive threshold algorithm. It extracts and fuses features from differential pressure, ultrasonic, and electrochemical signals. When the leakage probability exceeds a set threshold, the system generates a corresponding fault level and dynamically optimizes the fusion weights and threshold through reinforcement learning. The edge computing unit extracts features from differential pressure, ultrasonic, and electrochemical signals and inputs them into the multimodal fusion neural network for classification and identification, outputting three fault types: microbubble leakage, crack leakage, or material aging. The multimodal fusion neural network consists of three convolutional layers and two fully connected layers, with Softmax outputting the leakage type and an inference latency ≤50 ms. The three convolutional layers have 3×3 kernels with 32 / 64 / 128 channels; the two fully connected layers have 256 and 64 nodes respectively. The reinforcement learning within the edge computing unit employs Deep Q-Learning, dynamically adjusting the learning rate, fusion weights, and alarm thresholds of the convolutional neural network based on historical monitoring and maintenance records to continuously improve detection accuracy and early warning rate.
[0013] After the detection results are generated, the blockchain security evidence storage module hashes the event ID, timestamp, fault level, and key characteristic parameters, and uploads them to the blockchain storage to achieve tamper-proof and traceable management of the monitoring data. The blockchain security evidence storage module is configured with a smart contract that automatically triggers a maintenance work order when the same type of leakage is detected three times in a row, and uploads the work order ID and the original monitoring data to the blockchain at the same time.
[0014] The local alarm execution unit includes a buzzer, indicator lights, and an OLED screen, which outputs graded alarms according to the leakage level. The OLED screen supports touch operation, displays multimodal raw waveforms, fused feature maps, and predicted remaining life curves, and supports exporting CSV or image data via USB.
[0015] The multi-mode communication interface includes RS-485, Ethernet, LoRaWAN, or 5G communication modules, which are used to push JSON format early warning and diagnostic reports to SCADA systems, cloud platforms, and mobile terminals.
[0016] The self-calibration module utilizes the periodic injection of a known trace gas standard source and a reference liquid phase to achieve online automatic calibration of the sensor's zero point and sensitivity. Specifically, the self-calibration module automatically injects a standard gas source every 24 hours and performs joint calibration on the eight-channel differential pressure sensor, six circumferential ultrasonic transducers, and the polyaniline-modified carbon nanotube electrochemical sensing array, with a zero-point drift compensation accuracy ≤0.05%FS. The standard gas source is N2 + 50 ppm He.
[0017] The energy harvesting module includes a piezoelectric vibration energy harvester, a thermoelectric generator, and an energy storage module. The electrical energy generated by the piezoelectric vibration energy harvester and the thermoelectric generator is stored in the energy storage module, and the energy storage module provides auxiliary power to the eight-channel differential pressure sensor, six circumferential ultrasonic transducers, a polyaniline-modified carbon nanotube electrochemical sensing array, and the edge computing unit. The piezoelectric vibration energy harvester in the energy harvesting module is preferably a piezoelectric ceramic sheet, preferably 10mm×10mm×1mm in size, which can harvest power ≥5 mW from the ring vibration. The thermoelectric generator in the energy harvesting module is preferably a Bi2Te3-based thermoelectric generator, which outputs power ≥2 mW when the ring temperature difference is 5K~30K. The energy storage module is composed of solid-state capacitors, which are used to store the electrical energy generated by the energy harvesting module and provide backup power to the sensor and edge computing unit when the energy harvesting power fluctuates or is interrupted for a short time, with a continuous power supply time of not less than 20s.
[0018] The above technical solution has the following beneficial effects: 1. Multimodal fusion monitoring improves detection accuracy: By combining three types of sensors—differential pressure, ultrasonic, and electrochemical—this invention can comprehensively capture multi-source characteristic information of sealing ring leakage. Compared with a single sensing method, this invention can achieve differentiated detection of leakage and material aging, effectively reducing false alarms and missed alarms.
[0019] 2. High sensitivity and low latency response: The ultrasonic module, combined with standing wave peak and phase difference spectrum analysis, can detect microcracks down to 10µm with a response delay of ≤20 ms. The combination of multimodal fusion convolutional neural network and edge computing reduces the system inference delay to ≤50 ms, significantly improving detection sensitivity and real-time performance.
[0020] 3. Self-powered and low maintenance costs: The device integrates a piezoelectric vibration energy harvester and a Bi2Te3 thermoelectric generator, which can collect energy from the operating environment and store it in a solid capacitor to provide auxiliary power for sensors and edge computing units, reducing dependence on external power sources and maintenance costs, and improving the system's continuous operation capability.
[0021] 4. Online self-calibration ensures long-term stability: A standard gas source and reference liquid phase are set up, and the differential pressure, ultrasonic and electrochemical sensors are automatically injected and calibrated every 24 hours. The zero-point drift compensation accuracy is ≤0.05%FS, avoiding frequent manual calibration and ensuring the reliability and accuracy of long-term monitoring.
[0022] 5. Edge intelligence and adaptive optimization: The edge computing unit has a built-in deep reinforcement learning algorithm (Deep Q-Learning), which can dynamically adjust the neural network learning rate, fusion weights and thresholds based on historical monitoring and maintenance records, so as to continuously optimize the detection model and improve the ability to learn online for a long time and provide early warning.
[0023] 6. Blockchain secure storage ensures data trustworthiness: Event IDs, timestamps, fault levels, and key feature hashes are stored on the blockchain for evidence. The PBFT consensus mechanism is used to ensure that the monitoring logs are tamper-proof. When the same type of leakage occurs three times in a row, a maintenance work order is automatically triggered and stored on the blockchain, providing a secure and reliable basis for fault tracing and operation and maintenance decisions. Attached Figure Description
[0024] Figure 1 A flowchart for an online multimodal intelligent leak monitoring and alarm device; Figure 2 A schematic diagram showing the arrangement of an eight-channel differential pressure sensor and a circumferential ultrasonic transducer; Figure 3 This is a circuit block diagram of an electrochemical sensing array and energy harvesting module. Figure 4 This is a functional block diagram of the edge computing unit and the blockchain module. Detailed Implementation
[0025] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be further described in detail below with reference to the accompanying drawings.
[0026] Example 1 This embodiment provides an online multimodal intelligent leak monitoring and alarm device for turbine rubber sealing rings, such as... Figures 1-4 As shown, it includes an eight-channel differential pressure sensor, six circumferential ultrasonic transducers, a polyaniline-modified carbon nanotube electrochemical sensing array, an edge computing unit, a blockchain security storage module, a local alarm execution unit, a multi-mode communication interface, a self-calibration module, and an energy harvesting module.
[0027] This device is installed at the rubber sealing ring of the low-pressure cylinder of a 600 MW steam turbine. Eight differential pressure sensors are evenly arranged along the circumference of the sealing ring to collect the pressure difference signal between the inner and outer cavities of the sealing ring in real time. Six ultrasonic transducers are evenly installed along the circumference to collect ultrasonic signals generated by leakage. An electrochemical sensor array is installed in the steam sampling branch to detect electrochemical signals online.
[0028] The analog signals acquired by the eight-channel differential pressure sensor, six circumferential ultrasonic transducers, and polyaniline-modified carbon nanotube electrochemical sensing array are first digitized by the A / D module and then transmitted to the edge computing unit. The edge computing unit runs a multimodal fusion neural network based on the STM32H743 microprocessor and NPU coprocessor to perform feature extraction and fusion analysis on the differential pressure signal, ultrasonic signal, and electrochemical signal.
[0029] in: (1) Extract features such as average differential pressure and differential pressure fluctuation amplitude from differential pressure signals; (2) Extraction of standing wave peak amplitude and phase difference spectrum characteristics from ultrasonic signals; (3) Electrochemical signal extraction of H2O2 and Fe² + Concentration change rate characteristics.
[0030] Specifically, differential pressure signals were used to extract average differential pressure, peak-to-peak differential pressure, standard deviation of differential pressure, differential pressure non-uniformity between adjacent channels, and differential pressure change rate; ultrasonic signals were used to extract the standing wave peak amplitude change rate, envelope energy change rate, and characteristic frequency point phase shift; and electrochemical signals were used to extract H2O2 concentration and Fe²⁺ concentration. + Concentration and its rate of change relative to a reference state. In this embodiment, when the average pressure difference of any pressure difference channel changes more than 0.5 kPa relative to the reference state, or the peak-to-peak pressure difference is greater than 0.3 kPa, or the pressure difference non-uniformity between adjacent channels is greater than 1.0 kPa, the pressure difference signal is determined to be abnormal; when the rate of change of the peak amplitude of the ultrasonic standing wave is greater than 3%, and the rate of change of the envelope energy is greater than 5%, or the phase shift of the characteristic frequency point is greater than 5°, the ultrasonic signal is determined to be abnormal; when H2O2 or Fe²⁺ + An electrochemical signal is considered abnormal when the concentration increases by more than 20% relative to the baseline value and persists for more than 10 minutes.
[0031] The edge computing unit inputs the above features into a convolutional neural network for classification and identification, outputting three fault types: microbubble leakage, crack leakage, or material aging. When the leakage probability exceeds a set threshold, the system generates the corresponding fault level.
[0032] Meanwhile, the reinforcement learning module dynamically adjusts the feature fusion weights and alarm thresholds based on historical detection results and maintenance records to improve the detection accuracy of the system during long-term operation.
[0033] After the detection results are generated, the event ID, timestamp, fault level and key characteristic parameters are hashed and uploaded to the blockchain storage through the blockchain security storage module to achieve tamper-proof and traceable management of the monitoring data.
[0034] When the system determines that a level 1 or level 2 leakage fault has occurred, the local alarm execution unit will activate the buzzer and warning light to sound an alarm, and send alarm information to the SCADA system via RS-485, Ethernet or 5G communication module.
[0035] The energy harvesting module uses a piezoelectric vibration energy harvester to collect the vibration energy of the equipment and a Bi2Te3 thermoelectric generator to collect the surface temperature difference energy of the equipment. The electrical energy is stored in the energy storage module and then provides auxiliary power to the sensors and edge computing units.
[0036] In this embodiment, under leakage test conditions, the sealing ring exhibits a crack leakage of approximately 10μm. The system is able to identify the abnormal signal and trigger an alarm, thereby achieving online monitoring of the leakage status of the rubber sealing ring.
[0037] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.
Claims
1. An online multimodal intelligent leak monitoring and alarm device for rubber sealing rings of steam turbines, characterized in that: It includes an eight-channel differential pressure sensor, six circumferential ultrasonic transducers, a polyaniline-modified carbon nanotube electrochemical sensing array, an edge computing unit, a blockchain security storage module, a local alarm execution unit, a multi-mode communication interface, a self-calibration module, and an energy harvesting module. The eight-channel differential pressure sensors are evenly arranged at equal intervals along the circumference of the sealing ring to collect the pressure difference signal between the inner and outer cavities of the sealing ring in real time; the pressure difference signal includes the average pressure difference and the pressure difference fluctuation amplitude characteristics. The six circumferential ultrasonic transducers are uniformly installed along the circumference to collect ultrasonic signals generated by the leakage; the ultrasonic signals include standing wave peak amplitude and phase difference spectrum characteristics. The polyaniline-modified carbon nanotube electrochemical sensing array is configured in the steam sampling branch to detect electrochemical signals online; the electrochemical signals include H2O2 and Fe. 2+ Concentration change rate characteristics; The edge computing unit is based on the STM32H743 microprocessor and NPU coprocessor. It has a built-in multimodal fusion neural network and adaptive threshold algorithm to extract and fuse features of differential pressure signal, ultrasonic signal and electrochemical signal. When the leakage probability exceeds the set threshold, the system generates the corresponding fault level and dynamically optimizes the fusion weight and threshold through reinforcement learning. After the detection results are generated, the blockchain security storage module hashes the event ID, timestamp, fault level, and key characteristic parameters, and uploads them to the on-chain storage through the blockchain security storage module to achieve tamper-proof and traceable management of monitoring data. The local alarm execution unit includes a buzzer, indicator lights, and an OLED screen, and outputs graded alarms according to the leakage level. The multi-mode communication interface includes RS-485, Ethernet, LoRaWAN or 5G communication modules, which are used to push JSON format early warning and diagnostic reports to SCADA systems, cloud platforms and mobile terminals. The self-calibration module utilizes the periodic injection of a known trace gas standard source and a reference liquid phase to achieve online automatic calibration of the sensor's zero point and sensitivity; The energy harvesting module includes a piezoelectric vibration energy harvester, a thermoelectric generator, and an energy storage module. The electrical energy generated by the piezoelectric vibration energy harvester and the thermoelectric generator is stored in the energy storage module, and the energy storage module provides auxiliary power to the eight-channel differential pressure sensor, six circumferential ultrasonic transducers, polyaniline-modified carbon nanotube electrochemical sensing array, and edge computing unit.
2. The online multimodal intelligent leakage monitoring and alarm device for turbine rubber sealing rings according to claim 1, characterized in that: The six circumferential ultrasonic transducers emit pulsed ultrasonic signals and receive standing wave response signals from the sealing ring region. The ultrasonic signals generated by the leakage are acquired by extracting the standing wave peak amplitude change ΔA and phase difference spectrum characteristics.
3. The online multimodal intelligent leakage monitoring and alarm device for turbine rubber sealing rings according to claim 1, characterized in that: The polyaniline-modified carbon nanotube electrochemical sensing array detects electrochemical signals in the online manner as H2O2 and Fe²⁺ in steam. + The concentration change; the electrode area of the polyaniline-modified carbon nanotube electrochemical sensing array is 1 cm².
4. The online multimodal intelligent leakage monitoring and alarm device for turbine rubber sealing rings according to claim 1, characterized in that: The edge computing unit extracts differential pressure signal features, ultrasonic signal features, and electrochemical signal features and inputs them into a multimodal fusion neural network for classification and identification, outputting three fault types: microbubble leakage, crack leakage, or material aging.
5. The online multimodal intelligent leakage monitoring and alarm device for turbine rubber sealing rings according to claim 1, characterized in that: The multimodal fusion neural network consists of three convolutional layers and two fully connected layers, with Softmax output leakage type; the three convolutional layers are 3×3 convolutional kernels with 32 / 64 / 128 channels; the two fully connected layers have 256 and 64 nodes respectively.
6. The online multimodal intelligent leakage monitoring and alarm device for turbine rubber sealing rings according to claim 1, characterized in that: The reinforcement learning within the edge computing unit employs Deep Q-Learning, dynamically adjusting the learning rate, fusion weights, and alarm thresholds of the convolutional neural network based on historical monitoring and maintenance records to continuously improve detection accuracy and early warning rate.
7. The online multimodal intelligent leakage monitoring and alarm device for turbine rubber sealing rings according to claim 1, characterized in that: The blockchain security evidence storage module is configured with a smart contract that automatically triggers a maintenance work order when the same type of leakage is detected three times in a row, and uploads the work order ID and the original monitoring data to the blockchain at the same time.
8. The online multimodal intelligent leakage monitoring and alarm device for turbine rubber sealing rings according to claim 1, characterized in that: The OLED screen supports touch operation, displays multimodal raw waveforms, fused feature maps and predicted remaining lifetime curves, and supports exporting CSV or image data via USB.
9. The online multimodal intelligent leakage monitoring and alarm device for turbine rubber sealing rings according to claim 1, characterized in that: The self-calibration module automatically injects a standard gas source every 24 hours and performs joint calibration on the eight-channel differential pressure sensor, six circumferential ultrasonic transducers, and polyaniline-modified carbon nanotube electrochemical sensing array, with a zero-point drift compensation accuracy of ≤0.05%FS; the standard gas source is N2 + 50 ppm He.
10. The online multimodal intelligent leakage monitoring and alarm device for turbine rubber sealing rings according to claim 1, characterized in that: The piezoelectric vibration energy harvester is a piezoelectric ceramic plate with dimensions of 10mm×10mm×1mm, which can harvest power ≥5 mW from the vibration of the ring; the thermoelectric generator is a Bi2Te3 thermoelectric generator, which outputs power ≥2 mW when the temperature difference of the ring is 5K~30K; the energy storage module is composed of solid capacitors.
