Steam pipeline displacement monitoring system and method based on neuromorphic processing

Abstract

The invention provides a novel steam pipeline displacement monitoring system and method based on neuromorphic processing. The method comprises the following steps: installing a target and starting an active illumination module, collecting an event flow and an inertial measurement unit, performing time synchronization on metadata and working condition data, performing denoising and time plane construction on the event flow through a calculation processing module, and performing data processing on the event flow through a data processing module. The method comprises the steps of feature detection and event optical flow estimation, sub-pixel registration and pose solution, mapping pixel displacement into physical displacement, combining a state estimator and a pulse neural network, fusing event displacement observation and working condition data to obtain a dynamic allowable displacement envelope and an abnormal score, and triggering an alarm when a threshold value is exceeded and a duration time condition is met. And outputting a displacement curve and an alarm record, and finally carrying out recording and self-diagnosis. According to the invention, through fusion of neuromorphic processing and physical prior, accurate monitoring and abnormity alarm of steam pipeline displacement are realized, and the method has important application prospects and technical significance.

Description

Technical Field

[0001] This invention relates to the field of industrial steam pipeline equipment condition monitoring, and in particular to a steam pipeline displacement monitoring system and method based on neuromorphic processing. Background Technology

[0002] Steam pipelines are critical power transmission equipment in core industrial settings such as thermal power plants, petrochemical plants, and metallurgy. They operate under prolonged conditions of high temperature, high pressure, and alternating loads, making them prone to displacement and deformation due to uneven thermal expansion, accumulated stress on supports and hangers, and aging of pipeline materials. If the displacement exceeds a safe threshold, it can lead to pipeline joint seal failure, support and hanger detachment, or even pipeline rupture, causing major safety accidents such as steam leaks, fires, and explosions, while also resulting in energy waste and production interruptions. Therefore, real-time and accurate dynamic monitoring of steam pipeline displacement is a core requirement for ensuring the safe and stable operation of industrial systems and reducing maintenance costs.

[0003] Early steam pipeline displacement monitoring relied primarily on manual inspections combined with offline analysis, representing the industry's most basic monitoring model. Maintenance personnel were required to travel to the site periodically with specialized equipment. Mainstream testing tools included total stations, dial indicators, and vernier calipers. The specific implementation process was as follows: High-contrast metal targets were affixed or welded to critical parts of the pipeline as reference points for displacement measurement; when using a total station, a stable station was set up in an area far from the pipeline vibration source, and three-dimensional coordinate data was acquired by optically aiming at the target; when using a dial indicator, the base was fixed to a rigid support independent of the pipeline, ensuring close contact between the indicator and the pipeline surface to directly read minute displacement values; by comparing measurement data from different periods, and combining this with the "thermal expansion displacement threshold calculated based on material and operating conditions" in the pipeline design documents, if the actual displacement exceeded the threshold by 5%-10%, it was considered an abnormal displacement. This technology has significant limitations: severe lack of real-time performance, inability to capture instantaneous displacement caused by sudden pressure increases or temperature changes, poor environmental adaptability, high reliance on manual labor, and low efficiency.

[0004] Automated single-point sensor monitoring, based on different monitoring principles, can be divided into three main types: (1) Strain gauge displacement monitoring, in which metal foil strain gauges are glued to the outer wall of the pipe with high-temperature resistant adhesive. The strain gauges generate resistance changes as the pipe deforms. They are connected to a strain gauge via wires, and the resistance signal is converted into strain value. Then, the physical displacement is calculated based on the material mechanics formula. The data is transmitted to the monitoring system via RS485 wired transmission, and an automatic alarm is achieved by setting a strain threshold. (2) Laser displacement sensing monitoring, in which a laser displacement sensor is fixed on a rigid base away from the pipe. The sensor emits a laser beam to a reflective target on the surface of the pipe. The change in the distance between the sensor and the target is calculated by triangulation, and the displacement data is obtained. The data is transmitted to the backend via a LoRa / 4G wireless module. The sampling rate can reach 1kHz, and the response speed is fast. (3) Single-point monitoring of inertial measurement unit: A small inertial measurement unit module integrating accelerometer and gyroscope is fixed on the surface of the pipe by a high-temperature resistant clamp to collect the acceleration and angular velocity data of the pipe in real time, and obtain displacement information through integration calculation; the data can be stored locally or transmitted via Bluetooth at close range, without the need for external reference points, and is suitable for narrow spaces.

[0005] While these technologies achieve automated monitoring, they share common problems: incomplete spatial coverage, monitoring only localized points and failing to reflect the overall displacement distribution of the pipeline; insufficient environmental adaptability, with high temperatures and vibrations easily causing data distortion; and limited functionality, only outputting displacement values ​​and unable to correlate with operating conditions to determine whether the displacement is due to normal thermal expansion.

[0006] Visual inspection technology is gradually being applied to steam pipeline displacement monitoring, enabling simultaneous monitoring at multiple points. Based on technological evolution, it can be divided into traditional image processing technology and binocular stereo vision technology. Traditional image processing technology involves installing a visible light industrial camera next to the pipeline, with the lens aimed at key areas and targets. The camera periodically captures images at a frame rate of 1-10 fps. The backend uses a template matching algorithm to compare real-time images with a reference image, calculates changes in target pixel coordinates, and then converts the pixel displacement into physical displacement based on camera calibration parameters. Some systems are equipped with white LED supplementary lighting to cope with fluctuations in factory lighting. Binocular stereo vision technology uses two industrial cameras with a fixed distance to form a binocular system. It calculates the three-dimensional coordinates of the target using a stereo matching algorithm, solving the problem of depth measurement in traditional monocular vision. Periodic calibration of camera internal and external parameters ensures the accuracy of displacement calculation. While this technology expands the monitoring coverage, it still has key shortcomings: it cannot correlate with operating condition data, has limited environmental adaptability, can only determine the existence of displacement but cannot distinguish the severity of anomalies, and is difficult to determine a maintenance sequence.

[0007] Based on the aforementioned technological status, this invention proposes a steam pipeline displacement monitoring system and method based on neuromorphic processing. By integrating an event imaging module to address real-time performance and anti-interference issues, an active illumination module to ensure data validity in low-contrast and slow-displacement scenarios, an inertial measurement unit to compensate for vibration errors, and a fusion of temperature, pressure, and flow data from operating condition sensors, combined with high-precision clock synchronization, event stream preprocessing, and a fusion method of "physical prior (thermal expansion / compression model) + pulse neural network," this invention aims to overcome the core deficiencies of existing technologies, achieving full-domain coverage, accurate judgment, and anomaly alarm for steam pipeline displacement, providing reliable technical support for the safe operation and maintenance of industrial steam pipelines. Summary of the Invention

[0008] This invention proposes a steam pipeline displacement monitoring system and method based on neuromorphic processing. First, a high-contrast target is set up in the key monitoring area of ​​the steam pipeline, and an active illumination module is activated. An event imaging module collects the target brightness change event stream, while an inertial measurement unit synchronously collects pipeline attitude vibration data. Operating condition data is acquired through operating condition sensors. A clock synchronization module aligns the timestamps of the event stream, inertial data, and operating condition data. A computational processing module performs spatiotemporal consistency denoising and time surface construction on the event stream. After feature detection, event optical flow estimation, and sub-pixel registration, pixel displacement is mapped to physical displacement by combining camera calibration parameters and inertial data. Subsequently, the physical displacement and operating condition data form a state vector, which is input to a state estimator containing thermal expansion and pressure priors. A spiking neural network is used to complete data fusion, outputting a dynamic permissible displacement envelope and anomaly score. When a threshold is exceeded and a duration condition is met, an alarm is triggered through the output interface, generating a displacement curve and alarm record, ultimately achieving accurate monitoring and anomaly tracing of steam pipeline displacement.

[0009] This invention is based on the technical concept of "fusion of neuromorphic processing and physical priors." It leverages the high real-time performance of the event imaging module to address the latency issues of traditional visual detection, utilizes an inertial measurement unit to compensate for pipeline vibration interference, and employs an active illumination module to adapt to low-contrast and high-temperature conditions. Furthermore, it uses thermal expansion and pressure models as physical constraints to avoid misjudging normal displacement as abnormal. Ultimately, it adapts to the complex monitoring scenarios of steam pipelines characterized by high temperature, high pressure, frequent vibration, and fluctuating illumination, overcoming the shortcomings of existing technologies such as insufficient accuracy, weak environmental adaptability, and lack of correlation with operating conditions. The specific technical solution adopted is as follows:

[0010] The steam pipeline displacement monitoring system and method based on neuromorphic processing proposed in this invention have the following process: Figure 1As shown, its features are: 1) The computational processing module generates a dynamic displacement envelope using a thermal expansion / compression model as a physical prior, combined with a pulse neural network, self-learns to adapt to operating conditions, asynchronously fuses multi-source data, eliminates false signals, and solves the defect of high false alarm rate when operating conditions change in traditional algorithms, making anomaly detection more accurate. 2) The alarm uses dynamic envelope and anomaly scoring as joint criteria, the self-diagnostic subunit monitors the equipment status in real time and triggers maintenance, and the output supports multiple transmission methods and platform integration, balancing the accuracy of early warning and the convenience of operation and maintenance, avoiding the false alarms and passive operation and maintenance problems of traditional single threshold.

[0011] In this invention, the various parts of the system are described as follows:

[0012] 1. The event imaging module described in the patent refers to a single or two or more event cameras (selectably event-frame hybrid sensors (DAVIS type)), with multiple cameras forming a known baseline. Its main function is to acquire the event stream of brightness changes in high-contrast targets. Multi-camera configurations can improve displacement estimation accuracy through triangulation and adaptive uncertainty weighted fusion. Simultaneously, a stable world coordinate reference from the target is required to eliminate slow camera carrier drift. DAVIS-type sensors can briefly activate low frame rate frame assistance during periods of event sparseness or anomalous activity, enhancing monitoring robustness.

[0013] 2. The active illumination module described in the patent refers to a pulsed near-infrared light source with a center wavelength of 850nm or 940nm, a pulse frequency of 1-5kHz, and a duty cycle of 10%-30%. It achieves timing synchronization or gated exposure with the event imaging module via a trigger signal. Its main function is to induce a stable stream of brightness change events for the event imaging module in slow-movement or low-contrast scenes, ensuring the effectiveness of data acquisition.

[0014] 3. The optical and protective components described in the patent include a narrowband filter, a polarizer, a high-temperature resistant transparent protective window, an air curtain, and a heating and defogging sub-assembly. These optical and protective components ensure the stability of the acquired event stream.

[0015] 4. The inertial measurement unit described in the patent, in conjunction with the event imaging module, constitutes a vision-inertial estimator. Its main function is to collect equipment attitude and vibration data to compensate for line-of-sight jitter errors caused by camera carrier vibration, thereby improving the accuracy of displacement calculation.

[0016] 5. The operating condition data described in the patent refers to the temperature, pressure, and flow data obtained through operating condition sensors. Operating condition data directly reflects the operating environment and load status of the steam pipeline. Collecting this data provides environmental and operating condition background for displacement monitoring, reducing displacement measurement errors caused by fluctuations in operating conditions and making the monitoring results more consistent with the actual operating status of the pipeline.

[0017] 6. The computational processing module described in the patent refers to a processing unit that includes event stream spatiotemporal consistency denoising and time surface construction, feature detection and event optical flow estimation, sub-pixel registration, pose solving, data fusion and anomaly detection, and self-diagnosis.

[0018] 7. The feature detection and event optical flow estimation described in the patent employs event domain corner detection and time plane fitting optical flow estimation.

[0019] 8. The pose solution described in the patent is obtained by using robust loss (Huber or Cauchy) RANSAC.

[0020] The data fusion and anomaly detection described in Patent 9 are based on extended Kalman filtering or unscented Kalman filtering. The thermal expansion model and the simplified model of the pressure / support are used as process priors to generate a dynamic allowable displacement envelope, and the data is fused using a spiking neural network.

[0021] 10. The self-diagnosis described in the patent determines obstruction, window pollution, or lighting degradation by using event density, polarity entropy, contrast index, and time consistency index, and triggers air curtain / heating defogging control or maintenance prompts.

[0022] 11. The spiking neural network described in the patent refers to a network that uses pulse frequency encoding or time interval encoding and supports STDP or Hebbian online adaptive processing. Its main function is to asynchronously fuse event streams and operating condition data, adapt to normal patterns under different operating conditions through self-learning, assist in outputting anomaly scores, and improve the generalization of anomaly detection.

[0023] 12. The threshold described in the patent is specifically determined by using dynamic envelope deviation and anomaly score as joint criteria, and setting hysteresis and minimum duration to reduce false alarms.

[0024] The main features of this invention are as follows: This invention constructs a steam pipeline displacement monitoring system based on multi-source collaborative acquisition, neuromorphic fusion, and intelligent early warning. Its core features are reflected in three aspects: First, data acquisition breaks through the limitations of traditional single-module systems. Through the collaboration of event imaging, active lighting, inertial measurement, and operating condition sensors, it achieves three-dimensional coverage of displacement, vibration, and operating condition data, adapting to industrial scenarios. Second, computational processing integrates physical priors and neuromorphic processing technologies. Using thermal expansion / compression models as constraints, it generates dynamic displacement envelopes and combines them with a spiking neural network to asynchronously fuse multi-source data, solving the problems of high false alarm rates and poor operating condition adaptability in traditional algorithms. Third, the early warning and self-diagnosis mechanism uses a joint criterion of dynamic envelope and anomaly scoring to reduce false alarm rates, enabling real-time monitoring of equipment status and triggering maintenance, balancing accuracy and ease of operation and maintenance.

[0025] Benefits and application prospects of this invention: It can be applied to the field of industrial steam pipeline equipment condition monitoring to achieve real-time and accurate displacement of steam pipelines. Attached Figure Description

[0026] Figure 1 Flowchart of the present invention Detailed Implementation

[0027] like Figure 1 The diagram illustrates the complete workflow of a steam pipeline displacement monitoring system and method based on neuromorphic processing. This system leverages the asynchronous sensing characteristics of an event camera and the dynamic learning capabilities of a spiking neural network to achieve real-time monitoring and early warning of micro-displacements in steam pipelines under high temperature and high pressure environments.

[0028] In this embodiment of the invention, two DAVIS240C event-frame hybrid sensors of a certain model are used to construct the event imaging module. The core parameters are: resolution 346×260 (event mode) / 640×480 (frame mode), dynamic range 120dB, event trigger delay <1μs, and a baseline distance of 0.5m to form stereoscopic vision. High-contrast targets (checkerboard pattern, size 500×500mm) are installed at 10m intervals along key monitoring sections of the steam pipeline (elbows, flange connections, etc.). The sensors are fixed to shockproof brackets and equipped with double-layer quartz glass protective covers (IP67 protection level).

[0029] The active lighting module uses an 850nm pulsed near-infrared light source with the following parameters: pulse frequency adjustable from 1 to 5kHz, duty cycle 10% to 30%, and peak power 30W. It achieves time synchronization with the event camera via a TTL trigger signal (synchronization error < 50ns).

[0030] The inertial measurement unit selected is a certain model STIM380H, with the following parameters: gyroscope zero-bias stability 0.4. ° The system features a speed of / h, an accelerometer zero bias of 0.003mg, a sampling frequency of 200Hz, and a rigid connection with the event camera to form a visual-inertial tightly coupled system.

[0031] The operating condition sensors include three temperature (-50~300℃, accuracy ±0.5℃), three pressure (0-10MPa, accuracy ±0.2%FS), and three flow (0-500t / h, accuracy ±0.5%FS) sensors, which are deployed upstream and downstream of the pipeline monitoring section.

[0032] A fusion algorithm architecture combining spiking neural networks (SNNs) and extended Kalman filters (EKFs) is employed to achieve accurate displacement calculation and anomaly detection. The core algorithm includes: an event preprocessing layer using a spatiotemporal consistency denoising algorithm (based on a multi-resolution framework and temporal bilateral filtering) to eliminate ambient light interference events (retention rate > 95%); a feature extraction layer using an MV-Net network for event optical flow estimation, capturing the correlation between steep temporal changes and motion direction through a motion view transformation module; a pose calculation layer using a RANSAC algorithm with Huber loss (100 iterations, reprojection error threshold of 8 pixels) for robust registration; and a fusion decision layer using a pulse frequency encoded SNN network (containing 320 neurons in the input layer and 128 neurons in the hidden layer), supporting STDP learning rules to adapt online to changes in pipeline operating conditions. Through temporal alignment and feature fusion of event flow, inertial data, and operating parameters, the final output is the physical displacement (accuracy ± 0.1 mm) and anomaly score.

[0033] The following examples illustrate this:

[0034] S1. Data Acquisition: The event imaging module acquires the target brightness change event stream asynchronously, and the frame mode is automatically activated when events are sparse; the inertial measurement unit acquires six-axis attitude data; the operating condition sensor synchronously acquires temperature, pressure, and flow data, and the monitoring range covers the entire length of the pipeline and the surrounding 2m environmental area.

[0035] S2. Time Synchronization: A time synchronization network is constructed using the PTPv2 protocol (IEEE1588-2008). The industrial server is used as the reference clock to calibrate the time of event cameras, inertial measurement units and operating condition sensors, ensuring that the time deviation of multi-source data is ≤1ms.

[0036] S3. Data Processing and Fusion: Denoising and time-plane construction are performed on the event stream. Pixel displacement is mapped to physical displacement through feature detection, event optical flow estimation, sub-pixel registration, and pose solving. Temperature, pressure, and flow data are collected simultaneously to form a state vector. Combined with a state estimator and a spiking neural network, event displacement observations and operating condition data are fused to obtain a dynamic allowable displacement envelope and anomaly score.

[0037] S31. Denoising and temporal plane construction are performed on the event stream. Density statistics are performed on events within a 5×5 spatial window, and isolated events (spatial density < 3 events / pixel) are removed. Event polarity entropy is calculated using a sliding temporal window (window size 10mA). Entropy values ​​> 0.8 are identified as burst noise and filtered out. The event stream is accumulated into a temporal plane image (346×260×8bit) with 20ms intervals, preserving event polarity and timestamp information. Sparse regions are supplemented through interpolation.

[0038] S32. Feature extraction and optical flow estimation are performed. An improved FAST algorithm is used to extract target corner features from the temporal plane image (≥20 feature points per frame), and the spatiotemporal gradient histogram of the feature points is calculated as a descriptor. The continuous temporal plane image is processed by the MV-Net network to output a 16×16×2 dimensional optical flow field, and the pyramid LK algorithm is used to optimize the optical flow accuracy.

[0039] S33. Input the 3D coordinates of the corner points and the image coordinates into the RANSAC algorithm with Huber loss to solve for the camera extrinsic parameters. Combine the camera calibration parameters and baseline distance to convert the pixel displacement into physical displacement (unit: mm) and calculate the instantaneous velocity.

[0040] S34. The preprocessed displacement data and operating parameters are concatenated into a feature vector. The EKF process model incorporates the thermal expansion formula (ΔL=αL0AT) and a simplified compressive deformation model. Prior parameters are trained using 1000 hours of normal operating data, and the allowable displacement range (±3σ confidence interval) is updated every 5 minutes. The feature vector is input into the SNN via pulse frequency encoding, and the normal pattern is learned through STDP rules, outputting a 0-1 anomaly score (>0.7 triggers an early warning). The model is fine-tuned using 50 hours of abnormal data (including scenarios such as excessive displacement and vibration exceeding limits), improving the fusion accuracy by 12%.

[0041] S4. Based on the joint criterion of dynamic displacement envelope deviation and anomaly score, a graded early warning strategy is implemented, and the system availability is ensured through a self-diagnosis mechanism.

[0042] S41. A dual-parameter joint judgment rule is adopted: Level 1 warning: displacement deviation from the dynamic envelope is 1-3mm and the abnormality score is 0.7-0.8, with a duration of >30s. It is recommended to check the support status within 48 hours; Level 2 warning: displacement deviation is 3-5mm and the abnormality score is 0.8-0.9, with a duration of >10s. It is recommended to stop the machine for inspection within 24 hours; Level 3 warning: displacement deviation >5mm or abnormality score >0.9, with a duration of >5s. The warning mode is a red audible and visual alarm, requiring immediate action. All warning records include displacement curves, operating parameters, and event image snapshots, and the retention period is ≥2 years.

[0043] S42. The self-diagnostic subunit monitors the equipment's health status in real time, determining lens contamination or illumination degradation based on event density (<100 events / ms) and contrast index (<0.3), triggering air curtain cleaning (0.4MPa compressed air) or heating defogging (50℃ constant temperature). Monthly online model updates are performed, employing an active learning strategy to screen low-confidence samples for manual annotation, and compensating for seasonal temperature variations through domain adversarial training. The model retains nearly five iterations, supports rollback in abnormal scenarios, and ensures long-term monitoring accuracy ≥98%.

Claims

1. A novel steam pipeline displacement monitoring system and method based on neuromorphic processing, characterized in that: The event imaging module acquires the brightness change event stream of a high-contrast target on a steam pipeline. An active illumination module provides synchronous illumination to induce stable events. Optical and protective components ensure acquisition stability. An inertial measurement unit (IMU) acquires IMU data, and a condition sensor acquires condition data. The acquired event stream, IMU data, and condition data are synchronized in time. A computational processing module performs spatiotemporal consistency denoising on the event stream and constructs a time plane. Feature detection and event optical flow estimation are used to process the event stream to obtain target features and perform sub-pixel registration. Based on IMU data and camera calibration parameters, a pose solving method is used to map pixel displacement to physical displacement and velocity. Combining a state estimator that includes thermal expansion and pressure priors with a spiking neural network, data fusion and anomaly detection are performed on the event displacement observation and condition data to obtain a dynamic allowable displacement envelope and anomaly score. When the threshold is exceeded and the duration condition is met, an alarm is triggered, and the displacement curve and alarm record are output. Finally, recording and self-diagnosis are performed.

2. The steam pipeline displacement monitoring system and method based on neuromorphic processing according to claim 1, wherein the event imaging module refers to one or two or more event cameras (selectively event-frame hybrid sensors (DAVIS type)), with multiple cameras forming a known baseline. Its main function is to acquire the brightness change event stream of a high-contrast target. Multi-camera configuration can improve displacement estimation accuracy through triangulation and uncertainty adaptive weighted fusion. Simultaneously, a stable world coordinate reference from the target is needed to eliminate slow camera carrier drift. DAVIS-type sensors can briefly enable low frame rate frame assistance during periods of event sparseness or anomalies to enhance monitoring robustness.

3. In the steam pipeline displacement monitoring system and method based on neuromorphic processing according to claim 1, the active illumination module refers to a pulsed near-infrared light source with a center wavelength of 850nm or 940nm, a pulse frequency of 1-5kHz, and a duty cycle of 10%-30%, which is synchronized with the event imaging module in time or gated exposure via a trigger signal. Its main function is to induce a stable brightness change event stream for the event imaging module in slow displacement or low contrast scenes, ensuring the effectiveness of data acquisition.

4. The steam pipeline displacement monitoring system and method based on neuromorphic processing according to claim 1, wherein the optical and protective components include a narrowband filter, a polarizer, a high-temperature resistant transparent protective window, an air curtain, and a heating and demisting sub-assembly. The optical and protective components ensure the stability of the acquired event stream.

5. In the steam pipeline displacement monitoring system and method based on neuromorphic processing according to claim 1, the inertial measurement unit, in conjunction with the event imaging module, constitutes a vision-inertial estimator. Its main function is to collect equipment attitude and vibration data to compensate for line-of-sight jitter errors caused by camera carrier vibration, thereby improving the accuracy of displacement calculation.

6. The steam pipeline displacement monitoring system and method based on neuromorphic processing according to claim 1, wherein the operating condition data refers to the temperature, pressure, and flow data acquired through operating condition sensors. Operating condition data directly reflects the operating environment and load status of the steam pipeline. Collecting operating condition data provides environmental and operating condition background for displacement monitoring, reduces displacement measurement errors caused by operating condition fluctuations, and makes the monitoring results more consistent with the actual operating state of the pipeline.

7. The steam pipeline displacement monitoring system and method based on neuromorphic processing according to claim 1, wherein the computational processing module refers to a processing unit comprising event flow spatiotemporal consistency denoising and time surface construction, feature detection and event optical flow estimation, sub-pixel registration, pose solving, data fusion and anomaly detection and self-diagnosis.

8. The steam pipeline displacement monitoring system and method based on neuromorphic processing according to claims 1 and 7, wherein the feature detection and event optical flow estimation adopt event domain corner detection and time plane fitting optical flow estimation.

9. The steam pipeline displacement monitoring system and method based on neuromorphic processing according to claims 1 and 7, wherein the pose is solved by RANSAC with robust loss (Huber or Cauchy).

10. The steam pipeline displacement monitoring system and method based on neuromorphic processing according to claims 1 and 7, wherein the data fusion and anomaly discrimination are based on extended Kalman filtering or unscented Kalman filtering, using the thermal expansion model and the simplified pressure / support model as process priors to generate a dynamic allowable displacement envelope, and using a spiking neural network to fuse the data.

11. The steam pipeline displacement monitoring system and method based on neuromorphic processing according to claims 1 and 7, wherein the self-diagnosis is to determine obstruction, window contamination or lighting degradation by using event density, polarity entropy, contrast index and time consistency index, and trigger air curtain / heating defogging control or maintenance prompts.

12. The steam pipeline displacement monitoring system and method based on neuromorphic processing according to claims 1 and 10, wherein the spiking neural network refers to a network that uses pulse frequency encoding or time interval encoding and supports STDP or Hebbian online adaptive processing. Its main function is to asynchronously fuse event streams and operating condition data, adapt to normal patterns under different operating conditions through self-learning, assist in outputting anomaly scores, and improve the generalization of anomaly detection.

13. The steam pipeline displacement monitoring system and method based on neuromorphic processing according to claim 1, wherein the threshold is determined by using dynamic envelope deviation and anomaly score as joint criteria, and setting hysteresis and minimum duration to reduce false alarms.

Images