Real-time detection method, system and train for abnormality detection of roof high-voltage cable joints

Abstract

The present invention provides a method, system and train for real-time detection of abnormalities of high-voltage cable joints on the roof, which collects sound signals at high-voltage cable joints in real time; performs signal preprocessing, primary feature extraction and secondary feature extraction on the sound signals in sequence, wherein the primary feature extraction is to extract the primary features of each sub-segment of the sound signal in the time domain and frequency domain, and the secondary feature extraction is to extract the secondary features based on the extracted primary features by using statistical analysis; the classification model is called, the secondary features are classified by using the model, and the diagnosis results are output. Real-time abnormality detection can be achieved for the high-voltage cable joints on the roof.

Description

Technical Field

[0001] The invention belongs to the technical field of train fault diagnosis, and in particular relates to a method and system for real-time detection of abnormalities of a roof high-voltage cable joint, and a train. Background Art

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] Cable connectors are the main components of the power supply system. When they have abnormalities, they will affect the power supply of the vehicle and are related to the normal operation and safety of the train.

[0004] Currently, the detection of equipment failures mainly utilizes image recognition technology. However, when a high-voltage cable connector on the roof fails, there are situations where it cannot be detected simply through images. For example, if the internal insulation of the cable connector fails, a fault sound will be emitted, but it cannot be identified in the image.

[0005] In addition, the problems existing in the image recognition-based solution when used in cable joint fault diagnosis mainly include:

[0006] a) When a cable connector fails, such as when sparks occur, it is easily disturbed by sunlight, tunnel light, lights, etc., affecting the diagnostic effect;

[0007] b) There are few cable connector fault samples, resulting in poor image-based diagnosis results.

[0008] Based on sound signals, there are related acoustic imagers and other products both at home and abroad, which are used in places such as substations for fault diagnosis. However, such products can only diagnose faults such as insulation discharge under static conditions, and the equipment requires a sound array, which is not suitable for vehicle installation. Another anomaly detection method is the audio offline analysis and evaluation method, but due to the influence of factors such as the test location and distance of different testers, the ground offline analysis results are not ideal, and this method has certain safety risks.

[0009] In addition, the related models of sound abnormality recognition in the prior art are mainly implemented in two ways:

[0010] (1) Time-frequency domain threshold method: It uses a simple threshold to make judgments. However, this method sets the threshold based on experience, resulting in a high false alarm rate and inaccuracy.

[0011] (2) Deep learning method: This method requires a large number of fault samples for training. However, for the high-voltage cables on the roof, there are few fault samples, which affects the diagnostic effect of the deep learning model.

[0012] Therefore, the image recognition technology and related models of sound anomaly recognition in the prior art cannot be directly applied to real-time anomaly detection of vehicle-mounted high-voltage cable connectors. Summary of the invention

[0013] In order to solve the above problems, the present invention proposes a method, system and train for real-time detection of abnormalities of roof high-voltage cable joints. The present invention can realize real-time abnormality detection for roof high-voltage cable joints.

[0014] According to some embodiments, the present invention adopts the following technical solutions:

[0015] In the first aspect, a method for real-time detection of abnormalities in a high-voltage cable joint on a vehicle roof is disclosed, comprising:

[0016] Real-time collection of sound signals at high-voltage cable joints;

[0017] The high-voltage cable joint anomaly detection model is used to perform signal preprocessing, primary feature extraction, and secondary feature extraction on the sound signal in sequence. The primary feature extraction is to extract the primary features of each sub-segment of the sound signal in the time domain and frequency domain, and the secondary feature extraction is to extract the secondary features based on the extracted primary features using statistical analysis methods.

[0018] Call the high-voltage cable joint anomaly detection model, use the model to classify the secondary features, and output the diagnosis results.

[0019] As a further technical solution, the sound signal is preprocessed, specifically including:

[0020] Normalize the sound signal so that it can be processed at the same scale;

[0021] Take a data segment of a set length as an independent sample;

[0022] Filter the sound signal through high-pass filtering.

[0023] As a further technical solution, the sound signal is preprocessed, specifically including: extracting abnormal information of the high-voltage cable by minimum entropy deconvolution;

[0024] By transforming the filtered signal and obtaining the envelope, the low-frequency high-voltage cable abnormal signal is demodulated from the high-frequency signal.

[0025] As a further technical solution, the specific steps of performing a feature extraction on the preprocessed sound signal are:

[0026] Slice the audio samples of a set time length to obtain the features of each audio segment;

[0027] The primary features include time domain primary features and frequency domain primary features;

[0028] Time domain primary features: obtain the peak-to-peak value and kurtosis time domain features of each audio segment;

[0029] Frequency domain primary features: Perform Fourier transform on each audio segment to obtain frequency domain features.

[0030] As a further technical solution, when extracting secondary features, the primary features of all sub-segments are summarized to extract secondary features, which include: the maximum value, average value and variance of the primary features of each sub-segment in the time domain and frequency domain.

[0031] As a further technical solution, the training process of the high-voltage cable joint anomaly detection model is as follows:

[0032] Collect normal and abnormal historical data of the sound of the high-voltage cable joint on the roof;

[0033] Label historical audio to distinguish abnormal and normal audio;

[0034] Parse the audio file into a digital vector;

[0035] The digital vector is sequentially preprocessed, first feature extracted, and second feature extracted;

[0036] Extract samples of equal proportions, train the model, and obtain a trained machine learning classification model.

[0037] In the second aspect, a real-time detection system for abnormalities of high-voltage cable joints on a vehicle roof is disclosed, comprising:

[0038] The sound collection unit is installed at the high-voltage cable joint and is used to collect the sound signal at the high-voltage cable joint in real time;

[0039] A pre-processor, used for sampling the sound signal according to a set sampling frequency;

[0040] The detection host calls the anomaly detection model to perform anomaly detection on the sound signals collected in real time.

[0041] As a further technical solution, the anomaly detection model includes a preprocessing module, a primary feature extraction module, a secondary feature extraction module and a model application module;

[0042] The preprocessing module is configured to: perform signal preprocessing on the sound signal in sequence;

[0043] The primary feature extraction module is configured to: extract the primary features of each sub-segment of the sound signal in the time domain and the frequency domain;

[0044] The secondary feature extraction module is configured to: extract secondary features based on the extracted primary features using a statistical analysis method;

[0045] The model application module is configured to call the machine learning classification model to classify the secondary features and output the results.

[0046] As a further technical solution, the training process of the anomaly detection model is:

[0047] Collect normal and abnormal historical data of the sound of the high-voltage cable joint on the roof;

[0048] Label historical audio to distinguish abnormal and normal audio;

[0049] Parse the audio file into a digital vector;

[0050] The digital vector is sequentially preprocessed, first feature extracted, and second feature extracted;

[0051] Extract samples of equal proportions, train the model, and train the machine learning classification model.

[0052] In a third aspect, a train is disclosed, wherein a detection host is provided on the train, and a sound collection unit and a pre-processor are arranged at a high-voltage cable joint of the train body;

[0053] A sound collection unit, used for collecting sound signals at high-voltage cable joints in real time;

[0054] A pre-processor, used for sampling the sound signal according to a set sampling frequency;

[0055] The detection host calls the anomaly detection model to perform anomaly detection on the sound signals collected in real time.

[0056] Compared with the prior art, the present invention has the following beneficial effects:

[0057] The present invention proposes to pre-process the audio signal by deconvolution, thereby obtaining abnormal signals, and can realize the extraction of deep features of the audio signal based on a secondary extraction method.

[0058] The present invention can perform real-time diagnosis of abnormalities in the high-voltage cable joint on the roof through the above-mentioned abnormality detection model, and can predict faults before the cable joint fails completely based on sound.

[0059] As a real-time detection method, the present invention performs real-time cable fault diagnosis through sound signals, outputs fault probability values ​​in real time, then records and saves the fault conditions of the cables in the historical process (i.e., probability values), and analyzes the health status of the cables based on the changing trends of the probability values. Therefore, the health trends of high-voltage cables based on historical audio can be analyzed.

[0060] Advantages of additional aspects of the present invention will be given in part in the following description, and in part will become obvious from the following description, or will be learned through practice of the present invention.

[0061] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0063] Figure 1 It is a schematic diagram of a high-voltage cable joint detection system based on sound;

[0064] Figure 2 This is a schematic diagram of the training process of the abnormal detection model for high-voltage cable joints on the roof;

[0065] Figure 3 This is a schematic diagram of the real-time detection process of abnormalities in the high-voltage cable connector on the roof. DETAILED DESCRIPTION

[0066] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0067] It should be noted that the following detailed descriptions are all illustrative and intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0068] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprising" and / or "including" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0069] Embodiment 1:

[0070] In this embodiment, a high-voltage cable joint on the roof of a train is used as an example, but this does not mean that the real-time detection method or system provided by the present invention is only applicable to abnormal detection of high-voltage cable joints on the roof. It can also be applied to detection of high-voltage cable joints installed in other locations or in vehicles according to different installation locations or on-board terminals.

[0071] In this embodiment, a real-time abnormal detection system for a high-voltage cable joint on a roof is disclosed. Figure 1As shown, it includes: a sound sensor, a preprocessor, and a detection host.

[0072] Sound sensor: collects sound signals.

[0073] Preprocessor: Samples the sound signal, and the sampling frequency is set by the detection host.

[0074] Detection host: the environment in which the anomaly detection model is installed; on the one hand, it stores sound signals; on the other hand, it sets the sampling frequency of the preprocessor; and calls the anomaly detection model through the software environment to perform anomaly detection on the sound signals collected in real time.

[0075] Among them, the core module is the anomaly detection model, which includes a preprocessing module, a primary feature extraction module, a secondary feature extraction module and a model application module.

[0076] Preprocessing module: preprocess the collected sound signals, including:

[0077] Normalization: Normalize the sound signal so that the audio signal is processed at the same scale, and then perform sample decomposition after normalization.

[0078] Sample decomposition: Take a data segment with a duration of t0 as an independent sample.

[0079] Filtering: To avoid interference from low-frequency noise, the audio signal is filtered through high-pass filtering for the decomposed independent samples.

[0080] Minimum entropy deconvolution: It is a blind source separation method that extracts the pattern containing fault information from the many patterns obtained by decomposition. Since the sound signal of the high-voltage cable when it is abnormal will be modulated with the interference signal and noise signal, the abnormal information of the high-voltage cable is extracted by minimum entropy deconvolution of the filtered signal.

[0081] Envelope demodulation: By performing Hilbert transform on the filtered signal and obtaining the envelope, the low-frequency high-voltage cable abnormal signal is demodulated from the high-frequency signal.

[0082] It should be noted that minimum entropy deconvolution is equivalent to a filtering method, which filters out other interference signals in the sound signal. After filtering out the interference signals, the high-frequency signal remains.

[0083] In order to extract the low-frequency abnormal signal of the high-voltage cable, it is necessary to use envelope demodulation to extract the modulated low-frequency signal.

[0084] Therefore, minimum entropy deconvolution is a filtering method for high-frequency signals; envelope demodulation is a method for extracting low-frequency signals from high-frequency signals.

[0085] Through the above steps, the abnormal sound signals of the high-voltage cable can be filtered out.

[0086] Primary feature extraction module: Slice the audio sample of time length t0, each slice has a duration of t1, and obtain the primary features of each audio segment.

[0087] Time domain primary features: obtain the peak-to-peak value, kurtosis and other time domain features of each audio segment.

[0088] Frequency domain primary features: Perform Fourier transform on each audio segment to obtain frequency domain features.

[0089] Secondary feature extraction module: Since the abnormal signal of the high-voltage cable does not necessarily exist continuously, not every slice can reflect the abnormal information of the high-voltage cable. In addition, the abnormality of a certain sub-segment may be caused by random interference from the outside world; and the abnormality is judged by only a small audio segment, which affects the accuracy of the detection result. Therefore, based on the statistical analysis method, secondary features are extracted. Summarizing the primary features of all sub-segments and extracting secondary features can more accurately detect high-voltage cable abnormalities. Secondary features include: secondary features such as the maximum value, average value, and variance of the primary features in the time domain and frequency domain of each sub-segment.

[0090] Model application module: After the model is deployed, it performs preprocessing, primary and secondary feature extraction, machine learning classification model calling and result output according to the real-time input data and the method execution process.

[0091] In a specific implementation example, the machine learning classification method may adopt XGBoost, random forest, support vector machine, etc.

[0092] The above-mentioned scheme of the present disclosure arranges a sound sensor at the high-voltage cable joint to collect sound signals. The sound signals are sampled by a preprocessor and the digital signals are transmitted to the detection host. The sound signals are preprocessed by filtering, deconvolving, and finding the envelope of the sound signals. The time domain and frequency domain primary features are obtained by the fault frequency analysis method, and the secondary features are obtained based on the statistical analysis method. Based on the feature signals and sound labels, a sound anomaly detection model based on machine learning is trained. The roof high-voltage cable joint is detected for anomalies based on the model prediction results.

[0093] Since the fault sound is not emitted all the time, there is a situation where the sound is emitted intermittently. Therefore, this embodiment collects sound samples for a period of time, extracts a feature, slices the sample, extracts features for each section of the sample, and then performs feature statistics to extract secondary features. The above method can avoid the intermittent sound and affect the diagnosis effect. By diagnosing for a period of time through statistical analysis, rather than just diagnosing a point, the accuracy of abnormal diagnosis can be improved.

[0094] Based on the sound method, the present invention is more capable of predicting the fault before the cable joint fails completely, compared with the image method, which can only diagnose the fault when the fault is manifested. Therefore, the sound method can realize the fault prediction, and the image method is mainly used for fault diagnosis.

[0095] Embodiment 2:

[0096] Based on the system of embodiment 1, a method for real-time detection of abnormalities in a high-voltage cable joint on a roof is disclosed. Figure 3 As shown, specifically including:

[0097] Step 1: Real-time audio acquisition: Real-time audio data is acquired through the high-voltage cable joint detection system.

[0098] Step 2: Audio preprocessing: The audio signal is preprocessed through the preprocessing module.

[0099] Step 3: Primary feature extraction: Extract the primary features of each sub-segment in the time domain and frequency domain.

[0100] Step 4: Secondary feature extraction: Extract secondary features.

[0101] Step 5: Model call: Calculate the machine learning classification model results.

[0102] Step 6: Result output: If the model diagnosis result is abnormal, the abnormal information will be reported.

[0103] Through the above steps, real-time detection of abnormalities in the high-voltage cable joints on the roof can be achieved.

[0104] In this embodiment, the training process of the abnormal detection model for the roof high-voltage cable joint is as follows: Figure 2 shown.

[0105] Step 1: Historical audio data collection: Collect normal and abnormal historical data of the sound of the high-voltage cable connector on the roof.

[0106] Step 2: Sample labeling: Label historical audio to distinguish abnormal and normal audio.

[0107] Step 3: Audio file parsing: Since the historical audio is in wav or other audio formats, parse the audio file and convert it into a digital vector.

[0108] Step 4: Normalization: Normalize the audio digital vector.

[0109] Step 5: Sample decomposition: Truncate the audio into samples of t0 data length.

[0110] Step 6: High-pass filtering: Perform high-pass filtering on the audio signal.

[0111] Step 7: Minimum entropy deconvolution: extract abnormal information.

[0112] Step 8: Envelope demodulation: demodulate the low-frequency abnormal signal modulated on the high-frequency signal.

[0113] Step 9: Primary feature extraction: Extract the primary features of each sub-segment in the time domain and frequency domain.

[0114] Step 10: Secondary feature extraction: Extract secondary features.

[0115] Step 11: Sample extraction: Extract samples of equal proportion to train the model.

[0116] Step 12: Train the machine learning classification model.

[0117] Embodiment three:

[0118] In this implementation example, a train is disclosed, wherein a detection host is provided on the train, and a sound collection unit and a pre-processor are arranged at the high-voltage cable joint of the vehicle body;

[0119] A sound collection unit, used for collecting sound signals at high-voltage cable joints in real time;

[0120] A pre-processor, used for sampling the sound signal according to a set sampling frequency;

[0121] The detection host calls the anomaly detection model to perform anomaly detection on the sound signal collected in real time. For the specific detection process, please refer to the above-mentioned embodiment example 1 and embodiment example 2.

[0122] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

[0123] Although the above describes the specific implementation mode of the present invention in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art on the basis of the technical solution of the present invention without creative work are still within the scope of protection of the present invention.

Claims

1. A real-time detection method for abnormalities of high-voltage cable joints on a vehicle roof, characterized in that: include: Real-time collection of sound signals at high-voltage cable joints; The high-voltage cable joint anomaly detection model is used to perform signal preprocessing, primary feature extraction, and secondary feature extraction on the sound signal in sequence, wherein the primary feature extraction is to extract the primary features of each sub-segment of the sound signal in the time domain and frequency domain, and the secondary feature extraction is to extract the secondary features based on the extracted primary features using the statistical analysis method; the sound signal is preprocessed, specifically including: extracting the abnormal information of the high-voltage cable through minimum entropy deconvolution; By transforming the filtered signal and obtaining the envelope, the low-frequency high-voltage cable abnormal signal is demodulated from the high-frequency signal; When extracting secondary features, the primary features of all sub-segments are summarized to extract secondary features; Call the classification model, use the model to classify the secondary features, and output the diagnosis results.

2. The method for real-time abnormal detection of a roof high-voltage cable joint as claimed in claim 1, characterized in that: Perform signal preprocessing on the sound signal, including: Normalize the sound signal so that it can be processed at the same scale; Take a data segment of a set length as an independent sample; Filter the sound signal through high-pass filtering.

3. The method for real-time abnormal detection of a roof high-voltage cable joint as claimed in claim 1, characterized in that: The specific steps for extracting features from the preprocessed sound signal are: Slice the audio samples of a set time length to obtain the features of each audio segment; The primary features include time domain primary features and frequency domain primary features; Time domain primary features: obtain the peak-to-peak value and kurtosis time domain features of each audio segment; Frequency domain primary features: Perform Fourier transform on each audio segment to obtain frequency domain features.

4. The method for real-time abnormal detection of a roof high-voltage cable joint as claimed in claim 1, wherein the secondary features include: The maximum value, average value and variance of the primary features in the time domain and frequency domain of each sub-segment.

5. The method for real-time abnormal detection of a roof high-voltage cable joint as claimed in claim 1, characterized in that: The training process of the high-voltage cable joint anomaly detection model is as follows: Collect normal and abnormal historical data of the sound of the high-voltage cable joint on the roof; Label historical audio to distinguish abnormal and normal audio; Parse the audio file into a digital vector; The digital vector is sequentially preprocessed, first feature extracted, and second feature extracted; Extract samples of equal proportions, train the model, and train the machine learning classification model.

6. A real-time abnormal detection system for a high-voltage cable joint on a vehicle roof, using the real-time abnormal detection method for a high-voltage cable joint on a vehicle roof as claimed in claim 1, characterized in that: include: The sound collection unit is installed at the high-voltage cable joint and is used to collect the sound signal at the high-voltage cable joint in real time; A pre-processor, used for sampling the sound signal according to a set sampling frequency; The detection host calls the anomaly detection model to perform anomaly detection on the sound signals collected in real time.

7. The real-time abnormality detection system for the roof high-voltage cable joint as claimed in claim 6 is characterized in that: The anomaly detection model includes a preprocessing module, a primary feature extraction module, a secondary feature extraction module and a model application module; The preprocessing module is configured to: perform signal preprocessing on the sound signal in sequence; The primary feature extraction module is configured to: extract the primary features of each sub-segment of the sound signal in the time domain and the frequency domain; The secondary feature extraction module is configured to: extract secondary features based on the extracted primary features using a statistical analysis method; The model application module is configured to call the machine learning classification model to classify the secondary features and output the results.

8. The real-time abnormality detection system for the roof high-voltage cable joint as claimed in claim 6 is characterized in that: The training process of the anomaly detection model is: Collect normal and abnormal historical data of the sound of the high-voltage cable joint on the roof; Label historical audio to distinguish abnormal and normal audio; Parse the audio file into a digital vector; The digital vector is sequentially preprocessed, first feature extracted, and second feature extracted; Extract samples of equal proportions, train the model, and train the machine learning classification model.

9. A train, using the method for real-time abnormal detection of roof high-voltage cable joints as claimed in claim 1, characterized in that: The train is provided with a detection host, and a sound collection unit and a pre-processor are arranged at the high-voltage cable joint of the train body; A sound collection unit, used for collecting sound signals at high-voltage cable joints in real time; A pre-processor, used for sampling the sound signal according to a set sampling frequency; The detection host calls the anomaly detection model to perform anomaly detection on the sound signals collected in real time.

Images