Voiceprint recognition detection method for internal defects of drainage pipeline

Abstract

The invention belongs to the technical field of drainage pipeline detection, and particularly relates to a voiceprint recognition detection method for internal defects of a drainage pipeline, and the method comprises the following steps: S1, collecting sound signals in the pipeline through an acoustic sensor array; s2, preprocessing the collected sound signals, wherein the preprocessing comprises noise filtering, signal enhancement and segmentation processing; s3, extracting time domain, frequency domain and time-frequency domain features of the sound signals, and constructing voiceprint fingerprints; s4, establishing a defect classification model based on the deep neural network, and identifying different types of pipeline defects; s5, carrying out confidence evaluation on the detection result, and determining the position of the defect in the pipeline; s6, outputting a detection result and carrying out graded alarm; according to the method, various defect forms of cracks, blockage, damage and the like of different degrees of the drainage pipeline can be recognized, the recognition accuracy of the internal defects of the drainage pipeline can reach 95% or above through multi-dimensional voiceprint feature extraction and a deep learning algorithm, and continuous monitoring and real-time alarming can be achieved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of sewer pipeline detection, in particular to a voiceprint recognition method for detecting internal defects of a sewer pipeline. BACKGROUND

[0002] Traditional manual inspection is high in cost and low in efficiency, and has safety hazards, and cannot cover all pipe sections; pipeline robot detection equipment is expensive, has requirements for pipe diameter, and has limited mobility in complex pipe networks; and pipe endoscope detection is only applicable to directly accessible pipe sections, and has limited application range. Therefore, the present application proposes a voiceprint recognition technology-based method for detecting internal defects of a sewer pipeline. SUMMARY

[0003] In view of the limitations and a series of problems of traditional manual inspection, pipeline robot detection, and pipe endoscope detection, a voiceprint recognition technology-based method for detecting internal defects of a sewer pipeline is proposed, which realizes accurate identification and positioning of various defects such as cracks, blockages, and damage by analyzing the characteristics of sound signals in the pipeline, and solves the problems of low accuracy and poor applicability of existing detection methods.

[0004] To achieve the above-mentioned purpose, the present application provides the following technical scheme: A voiceprint recognition method for detecting internal defects of a sewer pipeline, comprising the following steps: S1: collecting sound signals in the pipeline through an acoustic sensor array; S2: pre-processing the collected sound signals, including noise filtering, signal enhancement, and segmentation processing; S3: extracting time domain, frequency domain, and time-frequency domain features of the sound signals to construct a voiceprint fingerprint; S4: establishing a defect classification model based on a deep neural network to identify different types of pipeline defects; S5: evaluating the confidence of the detection results and determining the location of the defects in the pipeline; S6: outputting the detection results and performing hierarchical alarm.

[0005] Preferably, the voiceprint feature extraction includes: Time domain features: signal amplitude, zero-crossing rate, and statistical features of short-time energy; Frequency domain features: power spectral density, main frequency component, and spectral centroid; Time-frequency domain features: wavelet transform coefficients and mel-frequency cepstral coefficients.

[0006] Preferably, the deep neural network structure is: input layer-hidden layer 1-hidden layer 2-output layer defect classification, which can identify normal, slight crack, severe crack, partial blockage, severe blockage, and pipe damage states.

[0007] Preferably, real-time feature extraction and defect judgment are performed on the collected sound signal, TDOA estimation based on GCC-PHAT is performed, the specific position of the defect in the pipeline is calculated, and the positioning robustness is improved through multi-pair sensor fusion and RANSAC de-extreme value, and decision support is provided based on the confidence score of the output defect type.

[0008] Preferably, the pipeline state, defect position and severity are displayed on the interface of the specific monitoring software, different levels of alarm mechanisms are set according to the defect severity, and finally the detection results are stored in the database to establish a pipeline health record.

[0009] Preferably, the pipeline health record: an independent health record is established for each section of the pipeline according to the inspection well interval, records the pipeline basic information, defect detection record in previous years, maintenance record, current health level, generates a health assessment report regularly, predicts the defect development trend, and guides preventive operation and maintenance.

[0010] Compared with the prior art, the beneficial effects of the present application are: 1、The present application can identify various defect forms of the drainage pipeline such as cracks, blockages and damages of different degrees, and through multi-dimensional voiceprint feature extraction and deep learning algorithm, the internal defect identification accuracy of the drainage pipeline can reach more than 95%.

[0011] 2、The present application adopts adaptive filtering and signal enhancement technology, which can effectively suppress the influence of environmental noise, can continuously monitor and real-time alarm, and can timely find the abnormality of the drainage pipeline.

[0012] 3、The detection method mentioned in the present application does not need to enter the inside of the pipeline, which avoids the limitation of the traditional manual detection method; compared with the robot detection method, the equipment and maintenance cost is lower. BRIEF DESCRIPTION OF DRAWINGS

[0013] Fig. 1 is a flow chart of the internal defect detection of the drainage pipeline of the present application; Fig. 2 is a data processing structure diagram of the present application. DETAILED DESCRIPTION

[0014] The technical solutions in the embodiments of the present application will be described clearly and completely in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0015] In the description of the present application, it should be understood that the terms "upper", "lower", "front", "back", "left", "right", "top", "bottom", "inner", "outer" and the like indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the present application and simplifying the description, and do not indicate or imply that the devices or elements referred to must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as a limitation on the present application.

[0016] Embodiments: Please refer to Figs. 1-2 The present application provides the following technical solutions: A drainage pipeline internal defect detection method based on voiceprint recognition technology, the specific implementation steps are as follows: S1, sound signal acquisition: high-precision acoustic sensor array needs to be deployed at the pipeline inspection well, multi-point synchronous acquisition method is adopted, and multiple types of acoustic information such as pipe water flow sound, bubble sound, leakage sound and structure impact sound are collected, and the sampling frequency can be generally set to 20kHz-50kHz, to ensure that the complete sound spectrum information is captured.

[0017] Deployment position: taking the drainage pipeline inspection well as the core deployment point, following the 30-50 meter interval principle (adjusting according to the pipe diameter: ≤1m interval 30m, >1m interval 50m); additional sensors are added at key nodes of the pipe network (such as turning places, slope change places, historical defect points), to ensure that the signal coverage has no blind area.

[0018] S2, signal preprocessing: the voiceprint data collected is preprocessed by using time-frequency signal analysis technology, including using an adaptive filtering algorithm to remove environmental noise and electromagnetic interference, amplifying and processing the weak abnormal sound signal for signal enhancement, and dividing the continuous signal into fixed-length data segments according to the time window.

[0019] The collected data is subjected to amplitude detection, and abnormal values exceeding the reasonable range (set to more than 3 times the normal signal amplitude, mostly caused by sensor instantaneous failure or external impact) are removed.

[0020] The segmented data segments need to meet the requirements of no broken frames and no overlapping redundancy, and each data segment is marked with the corresponding sensor number and timestamp.

[0021] S3, voiceprint feature extraction: the amplitude, zero-crossing rate, short-time energy and other time-frequency statistical features of the extracted signal are extracted, and the power spectrum density, main frequency component, spectral centroid and other frequency domain features are obtained through FFT transformation. Wavelet transform and mel frequency cepstrum coefficient (MFCC) can also be used to extract time-frequency features, and a voiceprint information library containing multi-dimensional feature vectors is constructed.

[0022] S4, Defect type classification model: a deep neural network is used to establish a defect recognition model. The design of the neural network needs to fully consider the characteristics of the voiceprint features. The network structure can be designed as input layer (120 dimensions) - hidden layer 1 (256 neurons) - hidden layer 2 (128 neurons) - output layer (defect type: normal, slight crack, serious crack, partial blockage, serious blockage, pipeline damage). ReLU and Softmax activation functions are used. The ReLU function is used in the hidden layer, which can effectively solve the gradient disappearance problem, accelerate the training convergence, and has the characteristics of sparsity. The Softmax function is used in the output layer, which converts the network output into a probability distribution, facilitating multi-classification decision-making. A cross-entropy loss function is used to handle the imbalance of defect categories.

[0023] S5, Model training and verification: 1000 or more sound samples of different defect types need to be collected. Each voiceprint sample is labeled and confirmed by a professional engineer. In special cases, training samples can be expanded by time shifting, noise addition, etc. Model parameters can be continuously optimized according to new samples collected during operation. 5-fold cross-validation is used to evaluate model performance. The model accuracy is ≥95%, and the recall rate is ≥90%.

[0024] Basic sample requirements: Collect sound samples of different pipe materials (concrete, plastic, cast iron) and different defect types, with a total of ≥1500 (original plan 1000 or more, further improve model generalization ability). Each sample is ≥5 seconds long. The labeled information includes defect type, pipe diameter, water flow rate, and environmental temperature.

[0025] Sample expansion strategy: In addition to time shifting, noise addition, amplitude scaling (±10%, simulating signal attenuation at different distances), frequency offset (±5%, simulating the speed difference of different material pipes), and time stretching (0.8-1.2 times, simulating signal changes at different water flow rates), the total number of expanded samples is ≥5000. The training set, validation set, and test set are divided in the ratio of 7:2:1.

[0026] Verification method: 5-fold cross-validation is used (the training set + validation set is divided into 5 parts, and 4 parts are used for training and 1 part is used for validation). This ensures the stability of the model performance. The test set (not involved in training) is used for final evaluation. The accuracy of the test set needs to be ≥94% to avoid model overfitting.

[0027] Model iteration: If the test set indicators do not meet the requirements, analyze the confusion matrix (such as severely confused defect types, supplement corresponding samples), re-optimize feature extraction or model structure until the indicators meet the requirements.

[0028] S6, Real-time detection and positioning: Real-time feature extraction and defect judgment on the collected sound signal, TDOA estimation based on GCC-PHAT, calculation of the specific position of the defect in the pipeline, and fusion of multiple pairs of sensors and RANSAC to improve the robustness of positioning, with positioning accuracy within ±2 meters, based on the confidence score of the output defect type, providing decision support.

[0029] Real-time feature extraction on the preprocessed signal, using a lightweight computing framework such as TensorRT to ensure that the feature extraction time is ≤50ms / segment; Input the feature vector into the trained classification model, output the defect type and confidence score (confidence ≥0.8 is judged as an effective detection result, <0.8 is marked as a suspected defect).

[0030] TDOA (Time Difference of Arrival) estimation based on GCC-PHAT (Generalized Cross Correlation-Phase Transform), calculating the time difference of the same defect signal arriving at different sensors, combining the sensor deployment location (latitude and longitude) and the pipeline sound speed (preset according to material: concrete pipe 3200m / s, plastic pipe 2800m / s, cast iron pipe 3500m / s), determining the defect location by triangulation.

[0031] Multi-sensor fusion: Select 3-5 pairs of sensors closest to the defect, calculate multiple sets of positioning results, and take the average as the preliminary positioning value; RANSAC outlier removal: RANSAC (Random Sample Consensus) algorithm processing on multiple sets of positioning results, removing outliers with error >5 meters (mostly caused by noise interference), and retaining positioning results with confidence ≥0.9; Position correction: Combined with the pipeline orientation (obtained from the GIS pipeline system), if the positioning result deviates from the pipeline axis by >3 meters, automatically correct it to the nearest point on the pipeline axis to ensure that the positioning is within the pipeline range.

[0032] S7, Result output and alarm: The pipeline state, defect location and severity can be displayed on the interface of a specific monitoring software, different levels of alarm mechanisms can be set according to the severity of the defect, and finally the detection results can be stored in the database to establish a pipeline health record.

[0033] Monitoring software interface design: Use three modules of map + list + chart for display; Map module: Overlay GIS pipeline map, mark defect location with different colors (green: normal, yellow: slight crack / partial blockage, orange: serious crack, red: serious blockage / pipeline damage), click on the defect point to view detailed information (defect type, confidence, positioning time, sensor number); List module: Sort by defect severity, show manhole number, defect type, positioning coordinates, confidence, processing status (unprocessed / processing / handled); Chart module: Statistics of defect type distribution (pie chart) and defect increase trend (line chart) in the past month, provide data support for operation and maintenance priority decision.

[0034] According to the defect severity, three levels of alarm are set: Level 1 alarm (emergency): serious blockage, pipe damage, trigger sound and light alarm + SMS notification + APP push (notification object: pipeline operation and maintenance person in charge, on-site repair personnel), require response within 1 hour and handle within 24 hours; Level 2 alarm (important): serious crack, trigger SMS notification + APP push (notification object: operation and maintenance person in charge), require response within 24 hours and handle within 72 hours; Level 3 alarm (general): slight crack, partial blockage, only APP push (notification object: operation and maintenance officer), require response within 1 week and handle within 1 month.

[0035] Alarm information includes defect location (latitude and longitude + manhole number), defect type, confidence, and recommended treatment plan.

[0036] Data storage: Store the detection results (original signal, preprocessed signal, feature vector, defect type, positioning result) to the cloud database (use MySQL+MongoDB hybrid architecture, structured data use MySQL, unstructured signal data use MongoDB), retention time ≥5 years, support historical data backtracking and analysis.

[0037] Pipeline health record: Establish an independent health record for each section of the pipeline (divided according to the interval between manholes), record the pipeline basic information (material, diameter, laying year), defect detection record, repair record, current health level (divided into excellent, good, medium, and poor according to the number and severity of defects), generate health assessment report regularly (every quarter), predict defect development trend, and guide preventive operation and maintenance.

[0038] The above shows and describes the basic principles and main features of the present application and the advantages of the present application. For those skilled in the art, it is obvious that the present application is not limited to the details of the above exemplary embodiments, and the present application can be realized in other specific forms without departing from the spirit or essential characteristics of the present application; therefore, from any point of view, the embodiments should be regarded as exemplary and non-limiting, the scope of the present application is defined by the appended claims rather than the above description, therefore all changes falling within the meaning and scope of the equivalent elements of the claims are intended to be included in the present application, any reference signs in the claims should not be considered as limiting the claims.

[0039] While embodiments of the application have been shown and described, it is to be understood that the embodiments described are merely exemplary of the principles and application of the present application. Numerous modifications and adaptions can be made without departing from the spirit and scope of the present application, which is defined by the following claims and their equivalents.

Claims

1. A method of acoustic fingerprinting for detecting defects in a sewer pipe, characterized in that, The method comprises the following steps: S1: Collecting sound signals in the pipeline through an acoustic sensor array; S2: Preprocessing the collected sound signals, including noise filtering, signal enhancement, and segmentation processing; S3: Extracting time-domain, frequency-domain, and time-frequency domain features of the sound signals to construct a voiceprint fingerprint; S4: Establishing a defect classification model based on a deep neural network to identify different types of pipeline defects; S5: Evaluating the confidence of the detection results and determining the location of the defects in the pipeline; S6: Outputting the detection results and issuing a graded alarm.

2. The acoustic fingerprinting method for detecting internal defects in a sewer pipe according to claim 1, characterized in that: The voiceprint feature extraction includes: Time-domain features: signal amplitude, zero-crossing rate, and statistical features of short-time energy; Frequency-domain features: power spectral density, main frequency component, and spectral centroid; Time-frequency domain features: wavelet transform coefficients and mel-frequency cepstral coefficients.

3. The acoustic fingerprinting method for detecting internal defects in a sewer pipe according to claim 1, characterized in that: The deep neural network structure is: input layer-hidden layer 1-hidden layer 2-output layer defect classification, which can identify normal, slight crack, severe crack, partial blockage, severe blockage, and pipeline damage states.

4. The acoustic fingerprinting method for detecting internal defects in a sewer pipe according to claim 1, characterized in that: Real-time feature extraction and defect judgment are performed on the collected sound signals, TDOA estimation based on GCC-PHAT is used to calculate the specific location of the defects in the pipeline, and the positioning robustness is improved through multi-pair sensor fusion and RANSAC de-extreme value, based on the confidence score of the output defect type, decision support is provided.

5. The acoustic fingerprinting method for detecting internal defects in a sewer pipe according to claim 1, characterized in that: The pipeline state, defect location, and severity are displayed on the interface of a specific monitoring software, an alarm mechanism of different levels is set according to the severity of the defects, and finally the detection results are stored in a database to establish a pipeline health record.

6. The acoustic fingerprinting method for detecting internal defects in a sewer pipe according to claim 5, characterized in that: Pipeline health record: an independent health record is established for each section of the pipeline according to the inspection well interval, the pipeline basic information, historical defect detection records, maintenance records, and current health level are recorded, a health assessment report is generated regularly to predict the defect development trend and guide preventive operation and maintenance.

Images