Real-time sensing method for third-party construction damage events

Abstract

The application provides a real-time sensing method for third-party construction damage events, and relates to the field of long-term monitoring of urban lifelines, which comprises the following steps: determining a cutoff frequency according to construction equipment information and an estimated noise power spectral density, and preprocessing an original vibration signal based on the cutoff frequency; calculating the product of the RMS values of P adjacent measuring points; combining D-S evidence theory to fuse the evidence of adjacent measuring points, eliminating abnormal interference signals by calculating a conflict coefficient K, and obtaining a processed signal; obtaining a dynamic threshold value, and determining the endpoints of the third-party construction damage event in the processed signal; and using a K nearest neighbor algorithm to perform feature matching and classification recognition on a to-be-identified sample, and identifying the event type of the third-party construction damage event. The application effectively improves the recognition accuracy and real-time performance, effectively eliminates abnormal data interference, enhances the robustness of feature extraction and classification, and significantly improves the intelligent level of the monitoring system.

Description

Technical Field

[0001] This invention relates to the field of long-term monitoring of urban lifelines, and more specifically, to a method for real-time sensing of third-party construction damage events. Background Technology

[0002] In the field of long-term monitoring of urban lifelines, especially in the health monitoring of critical municipal infrastructure such as underground pipelines, third-party construction damage incidents are one of the main external threats leading to structural damage and safety accidents. With the acceleration of urbanization, the density and complexity of underground pipelines are constantly increasing, and the potential damage risk to pipelines from third-party construction (such as excavator operations, rammed earth machine operations, and electric jackhammer operations) is also growing. Timely and accurate identification of third-party construction activities (such as excavator, rammed earth machine, and electric jackhammer operations) is of great significance for ensuring the safe operation of facilities and extending their service life.

[0003] Traditional third-party methods for monitoring construction damage incidents mainly rely on manual inspections or fixed-point sensors. These methods have significant limitations: Manual inspections are inefficient, costly, and difficult to implement 24 / 7 monitoring, easily missing important construction incidents. Fixed-point sensors are costly to deploy and have limited coverage, making them unsuitable for monitoring construction in large-area, complex environments.

[0004] In recent years, distributed optical fiber sensing (DAS) technology has been gradually applied to the monitoring of third-party sabotage events due to its advantages such as high sensitivity, full coverage and strong real-time performance. However, existing DAS-based monitoring methods still have the following problems.

[0005] The accuracy and real-time performance of the identification methods are insufficient. Most methods struggle to adapt to complex and variable construction interference signals and cannot achieve highly timely automatic identification; interference from abnormal data is a prominent issue. Environmental noise and sensor signal fluctuations easily interfere with the event identification process, and traditional methods struggle to effectively filter out abnormal data, leading to frequent false alarms and missed alarms. Traditional methods often use filters with fixed cutoff frequencies for signal preprocessing, which are ill-suited to complex and variable environmental noise, such as the nonlinear changes in noise spectrum caused by traffic flow variations and weather conditions. This results in the loss of effective information when construction signals and noise frequency bands overlap, increasing the risk of false alarms and missed alarms.

[0006] Furthermore, existing methods are highly dependent on feature extraction and classification modeling, have poor adaptability, and lack robustness and scalability. Therefore, there is an urgent need for a third-party construction damage event identification method that integrates multi-dimensional signal features, possesses adaptive capabilities, and has real-time perception capabilities to improve the intelligence level and response efficiency of infrastructure safety monitoring systems. Summary of the Invention

[0007] To overcome the aforementioned shortcomings of the prior art, the present invention is achieved through the following technical solution: A method for real-time sensing of third-party construction damage incidents, the method comprising: Pre-set initial parameters for signal acquisition in the distributed optical fiber sensing system, and based on the set initial parameters, acquire the original vibration signal generated by the optical fiber attached to the outer wall of the monitored pipeline. The construction equipment information is acquired in real time by a camera. The cutoff frequency is determined based on the construction equipment information and the estimated noise power spectral density. The original vibration signal is then preprocessed based on the cutoff frequency. Based on the calculation of the RMS value of each measuring point, the product of the RMS values ​​of P adjacent measuring points is calculated as evidence of adjacent measuring points, where P is the optimal number of adjacent measuring points; combined with the DS evidence theory, the evidence of adjacent measuring points is fused, and the abnormal interference signals in the preprocessed original vibration signal are removed by calculating the conflict coefficient K to obtain the processed signal; A dynamic threshold is obtained to determine the endpoints of third-party construction damage events in the processed signal. The endpoints include a start point and an end point. Based on the endpoints, valid signal segments are extracted from the processed signal, and samples to be identified are constructed. The K-nearest neighbor algorithm is used to perform feature matching and classification of the samples to be identified, thereby identifying the event type of third-party construction damage events.

[0008] Preferably, the initial parameters include sampling frequency, sampling range, number of blocks, and pulse width; wherein, the number of blocks is used to divide the entire sampling range into N consecutive measurement points.

[0009] Preferably, the step of acquiring construction equipment information in real time based on a camera and determining the cutoff frequency based on the construction equipment information and the estimated noise power spectral density includes: Real-time prediction of noise power spectral density (PSD) using a nonparametric method; Based on cameras installed within the protected area of ​​the pipeline being inspected, information on construction equipment is acquired in real time, including the type of construction equipment. Matching construction signal frequency bands based on the construction equipment information. ; The cutoff frequency at the current moment is determined based on the estimated noise power spectral density (PSD) and the construction signal frequency band.

[0010] Preferably, the step of preprocessing the original vibration signal based on the cutoff frequency includes: Design a digital high-pass filter based on the determined cutoff frequency at the current moment; The original vibration signal is taken as input and processed in real time by a digital filtering algorithm. The output signal of the digital filtering algorithm is the denoised signal.

[0011] Preferably, the steps of calculating the product of the RMS values ​​of P adjacent measuring points based on the RMS value of each measuring point, where P is the optimal number of adjacent measuring points; fusing the evidence from adjacent measuring points using DS evidence theory; and eliminating abnormal interference signals in the preprocessed original vibration signal by calculating the conflict coefficient K to obtain the processed signal include: The optimal number of adjacent measuring points P is determined based on the type of construction equipment. Calculate the RMS value for each measuring point, and use the energy product method to calculate the RMS product of P adjacent measuring points; By combining the DS evidence theory, the evidence from adjacent measuring points is fused, and the abnormal interference signals in the preprocessed original vibration signal are eliminated by calculating the conflict coefficient K, thus obtaining the processed signal.

[0012] Preferably, the step of determining the optimal number of adjacent measuring points P based on the type of construction equipment includes: Obtain a preset mapping table between device types and optimal P values, wherein the mapping table includes paired device types and values ​​of P; Based on the type of construction equipment acquired from the camera, the corresponding P value is selected from the preset mapping table between equipment type and optimal P value.

[0013] Preferably, the step of fusing evidence from adjacent measuring points using DS evidence theory and calculating the conflict coefficient K to remove abnormal interference signals from the preprocessed original vibration signal to obtain the processed signal includes: Construct a recognition framework and define a proposition set. ,in This indicates the proposition "An event occurs at this location," where the event refers to a third-party construction damage event. This indicates the proposition "No event occurred at this location"; Subsets of include: the empty set , , ; Basic trust assignment assigns a basic trust function to the evidence for each measurement point i. ; Using the Dempster synthesis rule, the basic trust assignment functions of P adjacent measurement points are sequentially fused. During the synthesis process, the conflict coefficient K is calculated; After fusion, a final confidence level is obtained by integrating evidence from all P measurement points. , , ;if If the conflict coefficient K is greater than the preset trust threshold and the conflict coefficient K is less than the first preset conflict threshold K1, then an event is confirmed to have occurred at that location; if the conflict coefficient K at a certain measuring point is greater than the second preset conflict threshold K2, then the signal at that point is an abnormal interference signal and is removed. After DS evidence theory fusion and conflict detection, the remaining signal is the processed signal, which is sent to subsequent steps for endpoint detection and classification.

[0014] Preferably, the steps of obtaining a dynamic threshold, determining the endpoints of third-party construction damage events in the processed signal, wherein the endpoints include a start point and an end point, extracting valid signal segments from the processed signal based on the endpoints, and constructing samples to be identified include: Acquire vibration signals within the target interval and calculate the average RMS value of all measuring points within the target interval as the dynamic threshold. Based on dynamic threshold scanning of the processed signal, the endpoints of third-party construction damage events in the processed signal are determined; Based on the endpoint, a valid signal segment is extracted from the processed signal, and the valid signal segment is cut into a small segment of preset duration as a sample to be processed.

[0015] Preferably, the step of determining the endpoints of third-party construction damage events in the processed signal based on a dynamic threshold scan includes: When the RMS value in the processed signal changes from below the dynamic threshold to continuously above the dynamic threshold, it is marked as the starting point, where continuous means that the duration exceeds T seconds. When the RMS value in the processed signal changes from being consistently above the dynamic threshold to being below the dynamic threshold, it is marked as a termination point.

[0016] Preferably, the step of using the K-nearest neighbor algorithm to perform feature matching and classification of the sample to be identified, and identifying the event type of the third-party construction damage event, includes: A sample set is constructed based on the original vibration signals at the time of historical third-party construction damage events, and the sample set includes a training set. For new samples to be identified, features are extracted using the same method as for the training set to obtain a feature vector. Calculate the distance between the feature vector of the sample to be identified and the feature vector of each sample in the training set; Based on the calculated distance, find the M closest samples to the sample to be identified from the training set; examine the labels of these M samples, and use the majority voting principle to select the event type that appears most frequently among the M labels as the final classification result of the sample to be identified.

[0017] Compared with the prior art, the beneficial effects of the present invention are: This invention effectively improves the accuracy and real-time performance of event identification. Specifically, it enhances the signal characteristics of spatially continuous events by calculating the product of RMS values ​​from adjacent measuring points, while suppressing random interference signals and improving signal validity. Furthermore, by combining DS evidence theory to fuse multi-point data and calculating conflict coefficients to eliminate abnormal interference signals, the accuracy of event identification is further improved. Moreover, by employing a digital high-pass filter and dynamic threshold scanning technology, real-time preprocessing and endpoint detection of the original vibration signal are achieved, meeting the requirements for high-efficiency monitoring.

[0018] This invention effectively eliminates interference from abnormal data. Specifically, it uses real-time construction equipment information acquired by cameras and combines it with the estimated noise power spectral density to determine the cutoff frequency. A digital high-pass filter is designed to preprocess the original vibration signal, effectively filtering out background noise and low-frequency interference. In addition, by calculating the conflict coefficient through DS evidence theory, the evidence weight of highly conflicting signals is automatically discounted, eliminating spatially isolated abnormal interference signals and reducing false alarms and missed alarms.

[0019] This invention enhances the robustness of feature extraction and classification by setting a dynamic threshold based on the average RMS value of the vibration signal within the target interval, adapting to signal changes under different construction scenarios and improving the accuracy of endpoint detection. It also employs the K-nearest neighbor algorithm to perform feature matching and classification of the samples to be identified, identifying the event type of third-party construction damage events, and determining the final classification result through majority voting, thereby enhancing the robustness and adaptability of the classification model.

[0020] This invention effectively improves the intelligence level of the monitoring system. It determines the optimal value of the number of adjacent measuring points P based on the type of construction equipment, optimizes the spatial fusion range, makes subsequent processing more targeted, and improves the recognition accuracy. It also utilizes distributed fiber optic sensing technology to achieve full-coverage monitoring, eliminating the need to deploy a large number of point sensors, reducing monitoring costs, and improving monitoring efficiency. Attached Figure Description

[0021] Figure 1 This is a flowchart of a real-time sensing method for third-party construction damage events provided in an embodiment of the present invention.

[0022] Figure 2 This is a comparison chart of the original signal waterfall plot and the processed signal waterfall plot in real time, as shown in the embodiment.

[0023] Figure 3 This example demonstrates the perception and classification effect of different construction damage events. Detailed Implementation

[0024] To make the technical problems to be solved, the technical solutions, and the beneficial effects of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the present invention and are not intended to limit the present invention.

[0025] Please see Figure 1 This invention provides a real-time sensing method for third-party construction damage events, the method comprising: Step S1: Set the initial parameters for signal acquisition in advance for the distributed optical fiber sensing system (DAS). Based on the set initial parameters, start and run the DAS system to acquire the original vibration signals generated by the optical fibers attached to or buried in the outer wall of the monitored pipeline. The initial parameters include sampling frequency, sampling range, number of blocks (N_points: dividing the entire sampling range into N consecutive measurement units (measurement points)), and pulse width; the sampling frequency is the number of times the signal is acquired at each measurement point per second, typically set to 500Hz, 1000Hz, or 2000Hz; the sampling range indicates which part of the fiber's length will be sampled, and its value represents the farthest distance that can be reached from the starting point of the fiber; simply select the sampling range where a valid signal exists according to actual needs; the number of blocks refers to how many times the fiber within the sampling range is divided. The sampling resolution is obtained by dividing the sampling range into N consecutive measurement units (also referred to as dividing the entire sampling range into N measurement points). The sampling resolution is the measurement range of each measurement point. If the sampling resolution is 10m, then the area measured by each sampling point is 10m. The pulse width is the pulse width of the probe pulse light in the optical fiber, which is generally selected as a value close to 10 times the sampling resolution. The original vibration signal is a one-dimensional spatiotemporal signal data containing third-party construction event signals and background noise. The optical fiber is connected to the distributed optical fiber sensing system (DAS) host to form a complete sensing loop. Step S2: Acquire construction equipment information in real time based on the camera, determine the cutoff frequency based on the construction equipment information and the estimated noise power spectral density, and preprocess the original vibration signal based on the cutoff frequency, i.e., filter and denoise. Background noise types include traffic noise, wind and rain noise, industrial noise, and noise from human activities. Traffic noise is mainly caused by the movement of vehicles such as cars and trucks, and is characterized by continuous low-frequency (usually concentrated in the range of 5-80Hz) broadband vibrations. Wind and rain noise is caused by raindrops hitting the ground or wind blowing against obstacles on the ground, and is characterized by high-frequency (>50Hz) random broadband vibrations. Industrial noise comes from factory machinery, water pumps, etc., and may contain specific line spectrum components (such as the 50Hz power frequency and its harmonics). Noise from human activities, such as pedestrians walking, is usually a sudden and irregular vibration. The types of incidents involving third-party construction damage include rammed earth, electric picks, and hammering; The steps to determine the cutoff frequency include: Step S211: Predict the noise power spectral density (PSD) in real time using a non-parametric method (such as the Welch method) instead of relying on fixed filter parameters, so that the system can sense and respond to noise changes caused by traffic flow, weather, etc. The specific steps include: Step S211a: Selecting the noise estimation signal segment. The system needs to intelligently select a "clean" or "most likely background noise" signal as the input for the Welch method; energy detection can be used to obtain the signal's short-time energy (or short-time RMS value) in real time. Signal segments with energy consistently below a certain low threshold are identified as "background noise segments" and extracted to obtain the noise estimation signal segment. This threshold can be preset based on the statistical characteristics of historical background noise data (such as mean + 2 standard deviations). Step S211b: Segmenting and windowing the signal segment. Specifically, the noise estimation signal segment x(n) of length T seconds selected in step S211a is divided into L overlapping short time periods (data windows). Each segment has N sampling points (e.g., 1024 points); the segments overlap by 50% (i.e., adjacent segments overlap by N / 2 points) to reduce the variance of the spectral estimation; each data segment is multiplied by a window function w(n) (e.g., Hanning window). The purpose is to reduce "spectral leakage" caused by the discontinuity of the beginning and end of the data segment; step S211c: perform Fast Fourier Transform (FFT) on each signal segment to obtain the frequency domain representation, further calculate the PSD estimate of each segment based on the frequency domain representation, and average the PSD estimates of all segments to obtain the final PSD. The Welch method for PSD estimation is an existing technology, and will not be described in detail here. Step S212: Based on the camera installed in the protected area of ​​the pipeline being inspected, real-time information on construction equipment is obtained. The construction equipment information includes the type of construction equipment, which includes a soil compactor and an electric pick. The specific steps include: using cameras installed within the protected area of ​​the pipeline being inspected, acquiring real-time construction images within the protected area, analyzing the images, and obtaining information about the construction equipment; the protected area refers to a certain range around the pipeline being inspected, which needs to be monitored closely to prevent damage from third-party construction. The protected area is primarily determined based on the importance and vulnerability of the pipeline, typically considering factors such as pipeline type, burial depth, surrounding environment, and historical damage records; analyzing the construction images to obtain information about the construction equipment usually employs image recognition technology, including but not limited to target detection algorithms (such as YOLO, SSD, etc.) and image classification algorithms. These algorithms identify the type of construction equipment in the images, such as rammed earth machines and electric picks, through training models; this is existing technology and will not be elaborated upon here. Step S213: Match the construction signal frequency band based on the construction equipment information. , This represents the lowest effective frequency (lower limit frequency) of the vibration signal from the construction equipment. Frequency below this level... The components are usually very weak in energy or mainly noise; This represents the highest effective frequency (upper limit frequency) of the vibration signal of the construction equipment. The specific steps are as follows: Based on the construction equipment type in the construction equipment information, query the construction signal frequency band of that construction equipment type from the event feature database. The steps for establishing the event feature database include: collecting a large amount of raw signal data of known types of third-party construction damage events as sample signals; performing time-frequency analysis on the collected sample signals to extract features of each type of third-party construction damage event, including but not limited to frequency domain features and time domain features. Frequency domain features include dominant frequency, centroid frequency, and frequency band energy distribution (such as the energy proportion of 0-50Hz and the energy proportion of 50-200Hz). Time domain features include signal amplitude, zero-crossing rate, and the period of the impact event; and storing the analyzed feature parameters into a database to obtain the event feature database. Step S214: Determine the cutoff frequency at the current moment based on the estimated noise power spectral density (PSD) and the construction signal frequency band. Specifically, determine the cutoff frequency at the current moment according to the cutoff frequency formula, which is shown below: ; in, The cutoff frequency at the current moment; It means The value obtained by multiplying the value by 95%; Traditional methods using fixed filtering parameters struggle to cope with complex and ever-changing environmental noise. This invention dynamically determines the optimal filter cutoff frequency by integrating real-time noise power spectral density (PSD) estimated using the Welch method with the type of construction equipment identified by the camera. This enables the system to automatically sense and respond to nonlinear changes in environmental noise caused by traffic flow, weather variations, etc., effectively solving the problem of lost signal information due to overlap between construction signals and noise frequency bands, and significantly reducing false alarm and missed alarm rates. The preprocessing of the original vibration signal based on the cutoff frequency includes: Step S221: Based on the determined cutoff frequency at the current time Design a digital high-pass filter (commonly an IIR filter such as a Butterworth or Chebyshev IIR filter, or an FIR filter). The characteristics of the digital high-pass filter are determined by its frequency response function. The complete characterization of this function defines the gain and phase delay characteristics of a digital high-pass filter for input signals of different frequencies, and its complex form is expressed as: ; in, It is the amplitude response, indicating the response of the digital high-pass filter to frequencies of . The gain (amplification) or attenuation (reduction) of the signal components; It is the phase response, indicating the response of the digital high-pass filter to a frequency of . The effect of phase delay on the signal components. The base of the natural logarithm, approximately 2.71828, is a special constant in mathematics used in exponential functions; It is a standard symbol in the fields of electrical engineering and signal processing. It represents the imaginary unit and is used to construct complex exponential functions that represent the phase characteristics of filters. The design goal of this step is to determine the system function of the filter (such as the transfer function or the coefficients of the difference equation) to achieve its amplitude response. Meets Qualcomm characteristics: Allows frequencies above the stated cutoff frequency Signal components pass through with high gain, while those below that gain are blocked. The signal components pass through. Specifically, the cutoff frequency. Typically defined as amplitude response Drop to passband reference gain The frequency point corresponding to approximately -3 dB; Step S222: The original vibration signal is used as input and processed in real time by a digital filtering algorithm (such as direct type I / II structure, recursive calculation using difference equation). Digital filtering algorithm is an existing technology and is widely used in the field of signal processing to achieve functions such as signal filtering and noise reduction. This invention will not elaborate on it in detail here. Step S223: The output signal of the digital filtering algorithm is the denoised signal. In this signal, frequencies lower than... The components (mainly corresponding to background noise) are due to the amplitude response. The value is very small and thus significantly suppressed, while values ​​higher than 100 are significantly suppressed. The components (including high-frequency features of construction events) are preserved, thus achieving the goal of noise suppression while retaining key event information.

[0026] Step S3: Based on the calculated RMS value of each measuring point, which is used to characterize the average energy of the signal at that measuring point, calculate the product of the RMS values ​​of P adjacent measuring points as evidence of adjacent measuring points, thereby enhancing the signal characteristics of each measuring point. This operation can amplify the energy characteristics of spatially continuous third-party construction event signals, while relatively suppressing the characteristics of spatially isolated random interference signals, where P is the optimal number of adjacent measuring points. Combining the DS evidence theory, the evidence from adjacent measuring points is fused, and the spatially isolated abnormal interference signals in the preprocessed original vibration signal are eliminated by calculating the conflict coefficient K, thereby improving the effectiveness of the third-party construction event signal and obtaining the processed signal. Step S301: Determine the optimal number of adjacent measuring points P based on the type of construction equipment. The specific steps are as follows: Obtain a preset mapping table between equipment types and optimal P values. The mapping table includes paired equipment types and P values. For example, a rammed earth machine has a wide vibration propagation range and excellent spatial continuity, so a larger P value, such as 5-7, is used to fully integrate multiple pieces of evidence and improve confidence. An electric pick has a concentrated vibration source and a relatively small influence range, so a smaller P value, such as 3-4, is used to avoid over-smoothing and loss of local features. Based on the type of construction equipment, the corresponding P value is selected from a preset mapping table between equipment types and optimal P values. This mechanism optimizes the spatial fusion range through adaptive adjustment of equipment types, making subsequent processing more targeted and improving recognition accuracy. This invention introduces an adaptive mechanism based on equipment type, dynamically selecting the optimal number of adjacent measuring points P according to the vibration propagation characteristics of different equipment such as rammed earth machines and electric picks. This not only makes the spatial fusion range more targeted, improving the confidence of monitoring large equipment and the ability to retain details in monitoring small equipment, but also avoids the waste of computing resources. The backend uses a lightweight K-Nearest Neighbors (KNN) algorithm for classification, ensuring high accuracy while meeting the real-time requirements of engineering applications for second-level response. Step S302: Calculate the RMS value for each measuring point. Based on this, calculate the RMS product of P adjacent measuring points using the energy product method; wherein, the measuring point... The formula for the RMS value is: ; in, For measuring points The RMS value is given by N, where N is the total number of measurement points. To improve sensitivity to events, the energy product method is used to calculate the RMS product of P adjacent measurement points (including i itself). That is, for each measurement point i, the RMS product of its neighboring points is calculated. The enhanced energy feature is obtained by multiplying the RMS values ​​of adjacent measurement points: ; Let i be the RMS product of its P adjacent measurement points; Step S303: Combining DS evidence theory, the evidence from adjacent measuring points is fused. By calculating the conflict coefficient K, spatially isolated abnormal interference signals in the preprocessed original vibration signal are eliminated, improving the effectiveness of the third-party construction event signal and obtaining the processed signal. This step involves each measuring point... The value is considered an independent piece of "evidence," and these spatial pieces of evidence are integrated to make a final decision. This process specifically includes the following steps: 1. Construct a recognition framework and define a proposition set. ,in This indicates the proposition "An event occurs at this location," where the event specifically refers to a third-party construction damage event. This indicates the proposition "No event occurred at this location"; Subsets of include: the empty set , , ; 2. Basic Trust Assignment (BBA), which assigns evidence (i.e., ...) to each measurement point i. Assigning basic trust function Specifically: to The value is mapped to the degree of confidence in proposition A. For example, Can be with The value is directly proportional to the value; at the same time, a portion of trust is allocated to "uncertainty". ,Right now ,satisfy: , This indicates the degree of confidence that the evidence from the i-th measurement point directly supports the proposition "an event occurred at this location"; This indicates the degree of confidence that the evidence from the i-th measurement point directly supports the statement "no event occurred at this location"; This represents the degree of confidence in the evidence from the i-th measurement point for the entire identification framework, i.e., the degree of uncertainty. 3. Using Dempster's synthesis rule, sequentially fuse the basic trust assignment functions of P adjacent measurement points. During the synthesis process, the conflict coefficient K is calculated. The K value quantifies the degree of conflict between evidence from different measurement points, and its calculation formula is as follows: (For all) The proposition is summed, where the value of K is between [0, 1]. It is a subset of propositions supported by evidence from the i-th test point. It is a subset of propositions supported by evidence from the z-th test point. Let represent the basic trust assignment function of the evidence from the i-th test point to proposition X, where X can be A, ¬A, or other subsets related to the entire recognition framework. It reflects the degree of direct trust the i-th test point has in the validity of proposition X based on its own signal characteristics (such as the calculated RMS value and other relevant information). Meaning and Similarly, z represents the z-th test point, and Y is also a proposition within the recognition framework. It represents the basic trust assignment function of the evidence from the z-th test point to proposition Y, reflecting the degree of trust the z-th test point has in the corresponding proposition based on its own signal characteristics. In the Dempster evidence theory fusion process, the basic trust assignment functions of different test points (such as the i-th and z-th test points) are fused. Through operations such as the Dempster synthesis rule, information from multiple test points is integrated to determine whether a third-party construction sabotage event has ultimately occurred. If the evidence from all P test points strongly supports proposition A, then the conflict between them is small, and the K value is close to 0. The fused result will support the event with extremely high confidence. If the evidence from a certain test point strongly supports A, but the evidence from its neighboring test points strongly supports ¬A (i.e., normal background), then the evidence is seriously contradictory to the surrounding environment, and the K value will be large. The DS rule will automatically discount the weight of such highly conflicting evidence during the synthesis process.

[0027] 4. After fusion, a final confidence level is obtained that integrates evidence from all P measurement points: , , ;if If the conflict coefficient K is greater than the preset confidence threshold and less than the first preset conflict threshold K1, then an event is confirmed to have occurred at that location, and the signal is considered valid. The preset confidence threshold is typically set to 0.5–0.7 based on the signal statistical characteristics of the construction scenario to balance sensitivity and robustness. If the conflict coefficient K of the fusion result at a certain measuring point is greater than the second preset conflict threshold K2, it indicates that the signal at that point is likely a spatially isolated anomalous interference signal (such as a random electromagnetic pulse or a slight impact at a single point) and should be discarded. The signal at that point will not be marked as an event. The threshold values ​​are 0 < first preset conflict threshold < second preset conflict threshold < 1. The first and second preset conflict thresholds are set according to actual usage requirements; typically, K1 is between 0.3 and 0.5, and K2 is between 0.6 and 0.8. K1 should be set lower than the minimum expected conflict value of a real event. For example, if K is typically ≤ 0.3 during large equipment operation, K1 can be set to 0.35 to ensure that high-confidence events are not mistakenly discarded. K2 should be set higher than the maximum expected conflict value of random interference. For example, if the K value for a single-point impact is typically ≥0.7, then K2 can be set to 0.65 to ensure that interference is accurately eliminated. In practical applications, the threshold can be determined by simulating different construction scenarios, statistically analyzing the distribution of K values ​​between real events and interference signals.

[0028] Step S304: After DS evidence theory fusion and conflict detection, the signals that are retained and confirmed to be generated by continuous events in real space are the processed signals with improved validity, and are sent to subsequent steps for endpoint detection and classification. This invention creatively combines spatial domain signal enhancement (RMS product of adjacent measurement points) with intelligent decision-making theory (DS evidence theory). By calculating the conflict coefficient K of multi-point evidence fusion, it can accurately identify and eliminate spatially isolated random anomalous interference (such as electromagnetic pulses and single-point impacts) while retaining continuous, real construction event signals. This elimination mechanism based on evidence conflict analysis is more scientific and reliable than the traditional single threshold method, greatly improving the effectiveness and reliability of the signal.

[0029] Step S4: Obtain a dynamic threshold to determine the endpoints of third-party construction damage events in the processed signal. The endpoints include a start point and an end point, i.e., the moment when the vibration signal begins to rise significantly from the background noise and the moment when it falls back to the background noise level. Based on these endpoints, extract valid signal segments from the processed signal and construct samples to be identified. The valid signal segments refer to the signal segments of third-party construction damage events. The samples to be identified are constructed based on these valid signal segments, specifically by cutting the valid signal segments into multiple short samples of fixed duration (e.g., 3 seconds). The training set is obtained by preparing samples using the method in step S4 and then labeling them with classification tags. The training set consists of many samples, each sample being a feature vector (an array of multiple features extracted from a 3-second signal, such as frequency domain energy and zero-crossing rate) and its corresponding label (e.g., the number 1). The label is a discrete category identifier, for example: 0: no event (background noise); 1: hammering; 2: rammed earth; 3: electric pick. Step S401: Acquire the vibration signal within the target interval, and calculate the average RMS value (or the average value plus several times the standard deviation) of all measuring points within the target interval as the dynamic threshold. The target interval refers to a signal interval containing only background noise and without any third-party construction damage events. In practice, this can be determined by observing indicators such as the signal's RMS value and spectral distribution, combined with historical data and experience. The average RMS value (or the average value plus several times the standard deviation) of all measuring points within the target interval is used as the dynamic threshold. Step S402: Scan the processed signal based on a dynamic threshold. When the RMS value in the processed signal changes from below the dynamic threshold to consistently above the dynamic threshold, it is marked as the starting point. "Consistent" refers to a duration exceeding T seconds; in this embodiment, T is 3 seconds. Continue scanning. When the RMS value of the signal changes from consistently above the dynamic threshold to below the dynamic threshold, it is marked as the ending point. The signal segment between the starting and ending points represents a potential event segment, i.e., a third-party construction damage event. Step S403: Extract effective signal segments from the processed signal based on the endpoints, and cut the effective signal segments into small segments of a preset duration as samples to be processed; For identifying third-party construction damage events, different types of such events typically have long durations and wide impacts. Therefore, in practical analysis, to address the small sample size challenge in machine learning, all data from the affected measurement points are extracted. The event duration for sample preparation is set at 3 seconds, yielding sample data for the machine learning classification model. This method transforms a single sample within a long time window into multiple samples within a short time window, increasing the sample size. Furthermore, since the consequences of violent construction in real-world engineering applications are generally severe, timely intervention is crucial after a third-party construction damage event. This places certain demands on the timeliness of data processing, classification, and alarms. The sample duration should not be too long during data analysis; therefore, this sample preparation method is reasonable.

[0030] Step S5: The K-nearest neighbor algorithm is used to perform feature matching and classification of the samples to be identified, thereby identifying the event type of third-party construction damage events and realizing real-time identification of different types of third-party construction damage events. The specific steps are as follows: Based on the original vibration signals from historical third-party construction damage events, a sample set is constructed, which includes a training set. Features of new samples to be identified are extracted using the same method as the training set, resulting in a feature vector. The distance between the feature vector of the sample to be identified and the feature vector of each sample in the training set is calculated (usually using Euclidean distance). Based on the calculated distance, the M closest samples (M is a preset integer) to the sample to be identified are found from the training set. The labels of these M samples are examined, and a majority vote is used. The event type that appears most frequently among these M labels is taken as the final classification result of the sample to be identified. Specifically, the construction of the training set based on the original vibration signals from historical third-party construction damage events includes: acquiring original vibration signal segments from historical third-party construction damage events; cutting the original vibration signal segments into multiple short signal segments; extracting feature vectors for the samples (i.e., calculating a series of mathematical features for each segmented signal segment); labeling each feature vector with a corresponding label; and extracting the labeled feature vectors to form a sample set, which includes the training set. The label here represents the event type of the third-party construction damage event corresponding to the signal segment, including hammering, tamping, electric picks, etc. To ensure uniform sample length and suitability for model input, the original vibration signal segments from each historical third-party construction damage event were segmented into multiple 3-second segments. If an event lasted 9 seconds, it would generate three 3-second samples. The feature vectors extracted from these 3-second signal segments and their corresponding labels were used as samples for the machine learning classification model. The collection of all samples constituted the sample set used for training and testing the classification model.

[0031] Feature vector extraction of a sample involves calculating a series of mathematical features for each segmented signal segment. All calculated mathematical features are then sequentially combined into a one-dimensional array, which represents the feature vector of a sample. Because the value ranges of different mathematical features vary significantly (e.g., the mean might be 0.1, while the band energy might be 1000), directly using them in the K-NN algorithm would lead to features with larger value ranges dominating the results. Therefore, it is necessary to standardize all feature vectors in the entire training set, commonly using Z-score standardization. The mathematical features include time-domain features (describing the statistical properties of signal amplitude changes over time) and frequency-domain features (transforming the signal from the time domain to the frequency domain to analyze its frequency components). Time-domain features include mean, root mean square, amplitude, and zero-crossing rate. Frequency-domain features include spectral centroid, spectral bandwidth, and band energy.

[0032] Based on the above steps, real-time detection of third-party construction damage incidents can be achieved.

[0033] The following is an example of data collected by a DAS system over a certain period of time, used to identify third-party construction damage events using the methods described above. Please refer to [link / reference]. Figures 1 to 3 .

[0034] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0035] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

[0036] Those skilled in the art will also understand that the various exemplary logic blocks, modules, circuits, and algorithm steps described in connection with the present application can be implemented as electronic hardware, computer software, or a combination of both.

[0037] The flowcharts in the accompanying drawings illustrate the system operation of possible implementations of the methods according to various embodiments of this application. In this regard, each block in the flowchart may represent a module, segment, or portion of code containing one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than that shown in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved.

[0038] The various embodiments of this application have been described above. These descriptions are exemplary and not exhaustive, nor are they limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is chosen to best explain the principles, practical application, or improvement of the technology in the market, or to enable others skilled in the art to understand the embodiments disclosed herein.

Claims

1. A method for real-time sensing of third-party construction damage events, characterized in that, The method includes: Pre-set initial parameters for signal acquisition in the distributed optical fiber sensing system, and based on the set initial parameters, acquire the original vibration signal generated by the optical fiber attached to the outer wall of the monitored pipeline. The construction equipment information is acquired in real time by a camera. The cutoff frequency is determined based on the construction equipment information and the estimated noise power spectral density. The original vibration signal is then preprocessed based on the cutoff frequency. Based on the calculation of the RMS value of each measuring point, the product of the RMS values ​​of P adjacent measuring points is calculated as evidence of adjacent measuring points, where P is the optimal number of adjacent measuring points; combined with the DS evidence theory, the evidence of adjacent measuring points is fused, and the abnormal interference signals in the preprocessed original vibration signal are removed by calculating the conflict coefficient K to obtain the processed signal; A dynamic threshold is obtained to determine the endpoints of third-party construction damage events in the processed signal. The endpoints include a start point and an end point. Based on the endpoints, valid signal segments are extracted from the processed signal, and samples to be identified are constructed. The K-nearest neighbor algorithm is used to perform feature matching and classification of the samples to be identified, and to identify the event type of third-party construction damage events; The step of acquiring construction equipment information in real time based on a camera and determining the cutoff frequency based on the construction equipment information and the estimated noise power spectral density includes: Real-time prediction of noise power spectral density (PSD) using a nonparametric method; Based on cameras installed within the protected area of ​​the monitored pipeline, construction equipment information is acquired in real time, including the type of construction equipment. Matching construction signal frequency bands based on the construction equipment information. ; Based on the estimated noise power spectral density (PSD) and the construction signal frequency band, determine the cutoff frequency at the current moment; The steps of calculating the RMS value of each measuring point, multiplying the RMS values ​​of P adjacent measuring points as evidence of adjacent measuring points (where P is the optimal number of adjacent measuring points), and fusing the evidence from adjacent measuring points using DS evidence theory, and calculating the conflict coefficient K to remove abnormal interference signals from the preprocessed original vibration signal to obtain the processed signal include: The optimal number of adjacent measuring points P is determined based on the type of construction equipment. Calculate the RMS value for each measuring point, and use the energy product method to calculate the RMS product of P adjacent measuring points as evidence of adjacent measuring points; By combining the DS evidence theory, the evidence from adjacent measuring points is fused, and the abnormal interference signals in the preprocessed original vibration signal are eliminated by calculating the conflict coefficient K, thus obtaining the processed signal.

2. The real-time sensing method for third-party construction damage events according to claim 1, characterized in that, The initial parameters include sampling frequency, sampling range, number of blocks, and pulse width; wherein, the number of blocks is used to divide the entire sampling range into N consecutive measurement points.

3. The real-time sensing method for third-party construction damage events according to claim 1, characterized in that, The step of preprocessing the original vibration signal based on the cutoff frequency includes: Design a digital high-pass filter based on the determined cutoff frequency at the current moment; The original vibration signal is taken as input and processed in real time by a digital filtering algorithm. The output signal of the digital filtering algorithm is the denoised signal.

4. The real-time sensing method for third-party construction damage events according to claim 1, characterized in that, The step of determining the optimal number of adjacent measuring points P based on the type of construction equipment includes: Obtain a preset mapping table between device types and optimal P values, wherein the mapping table includes paired device types and values ​​of P; Based on the type of construction equipment acquired from the camera, the corresponding P value is selected from the preset mapping table between equipment type and optimal P value.

5. The real-time sensing method for third-party construction damage events according to claim 1, characterized in that, The steps of fusing evidence from adjacent measuring points by combining DS evidence theory and eliminating abnormal interference signals in the preprocessed original vibration signal by calculating the conflict coefficient K to obtain the processed signal include: Construct a recognition framework and define a proposition set. ,in This indicates the proposition "An event occurs at this location," where the event refers to a third-party construction damage event. This indicates the proposition "No event occurred at this location"; Subsets of include: the empty set , , ; Basic trust assignment: Assigning a basic trust function to the evidence for each measurement point i. ; Using the Dempster synthesis rule, the basic trust assignment functions of P adjacent measurement points are sequentially fused. During the synthesis process, the conflict coefficient K is calculated; After fusion, a final confidence level is obtained by integrating evidence from all P measurement points. , , ;if If the conflict coefficient K is greater than the preset trust threshold and the conflict coefficient K is less than the first preset conflict threshold K1, then an event is confirmed to have occurred at that location; if the conflict coefficient K at a certain measuring point is greater than the second preset conflict threshold K2, then the signal at that point is an abnormal interference signal and is removed. After DS evidence theory fusion and conflict detection, the remaining signal is the processed signal, which is sent to subsequent steps for endpoint detection and classification.

6. The real-time sensing method for third-party construction damage events according to claim 1, characterized in that, The steps of obtaining a dynamic threshold, determining the endpoints of third-party construction damage events in the processed signal (the endpoints including a start point and an end point), extracting valid signal segments from the processed signal based on the endpoints, and constructing samples to be identified include: Acquire vibration signals within the target interval and calculate the average RMS value of all measuring points within the target interval as the dynamic threshold. Based on dynamic threshold scanning of the processed signal, the endpoints of third-party construction damage events in the processed signal are determined; Based on the endpoint, a valid signal segment is extracted from the processed signal, and the valid signal segment is cut into a small segment of preset duration as a sample to be identified.

7. The real-time sensing method for third-party construction damage events according to claim 6, characterized in that, The step of determining the endpoints of third-party construction damage events in the processed signal based on dynamic threshold scanning includes: When the RMS value in the processed signal changes from below the dynamic threshold to continuously above the dynamic threshold, it is marked as the starting point, where continuous means that the duration exceeds T seconds. When the RMS value in the processed signal changes from being consistently above the dynamic threshold to being below the dynamic threshold, it is marked as a termination point.

8. The real-time sensing method for third-party construction damage events according to claim 1, characterized in that, The steps of using the K-nearest neighbor algorithm to perform feature matching and classification of the samples to be identified, and to identify the event type of third-party construction damage events, include: A sample set is constructed based on the original vibration signals from historical third-party construction damage events, including a training set. For new samples to be identified, features are extracted using the same method as for the training set to obtain a feature vector. Calculate the distance between the feature vector of the sample to be identified and the feature vector of each sample in the training set; Based on the calculated distance, find the M closest samples to the sample to be identified from the training set; examine the labels of the M samples, and use the majority voting principle to select the event type that appears most frequently among the M labels as the final classification result of the sample to be identified.

Images