Intelligent inspection method and system

Through intelligent inspection methods, sensors are used to obtain the real-time state of the electromechanical pipeline, combined with wavelet transformation and deep learning models, comprehensive monitoring of the pipeline is achieved, solving the problems of single detection and unsolved safety hazards in the existing technology, and improving the pipeline safety inspection accuracy and comprehensiveness.

CN118942171AInactive Publication Date: 2024-11-12JIANGXI TULAN METAL TECH CO LTD
View PDF 7 Cites 0 Cited by

Patent Information

Application Number
CN202411435114.9
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-10-15
Publication Date
2024-11-12
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

The prior art can only simply check the outer surface and temperature when detecting building electromechanical pipelines, and cannot effectively detect safety hazards and damage of the pipelines. The detection is single, so the pipeline situation cannot be fully monitored.

Method used

Using intelligent patrol methods, the real-time state of the electromechanical pipeline is obtained through the sensors on the support bracket, and converted into vibration signal text, image signal text and temperature signal text, combining wavelet transformation, preset decomposition, double-layer sliding window algorithm and YOLO, Fast RCNN network model, performs time-frequency graph feature analysis and image segmentation, calibrates the suspected area, and monitors pipeline disturbances, damage and temperature safety status.

Benefits of technology

It realizes comprehensive monitoring of electromechanical pipelines, can judge whether the pipeline is operating normally, detect surface damage, and evaluate the temperature safety status of each area, improving the accuracy and comprehensiveness of pipeline safety inspection.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118942171A_ABST
    Figure CN118942171A_ABST
Patent Text Reader

Abstract

The invention provides an intelligent inspection method and system, and the method comprises the steps: obtaining a real-time state of an electromechanical pipeline, and converting the real-time state into a state text; processing the plurality of wavelet high-frequency coefficients based on wavelet inverse transformation to obtain a reconstructed de-noised signal; transforming the reconstructed de-noised signal to obtain a time-frequency diagram, calculating time-frequency domain characteristic parameters of the time-frequency diagram, and marking the time-frequency domain characteristic parameters by adopting a threshold value combined with a double-layer sliding window algorithm; segmenting the image signal text into a plurality of grid images to obtain a prediction image; a suspected area in the prediction image is calibrated, and a suspected target in the suspected area is obtained through mapping; and comparing the temperature information of the plurality of grid images with a safety threshold temperature. According to the invention, the temperature, the appearance and the interior of the electromechanical pipeline can be monitored, and the temperature safety state of each area of the electromechanical pipeline can be monitored by combining the image signal text with the temperature signal text.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of building monitoring, and in particular to an intelligent inspection method and system. Background Art

[0002] There are numerous building mechanical and electrical engineering projects, a large number of mechanical and electrical equipment, and complex pipelines. Building mechanical and electrical pipeline facilities are non-structural components. When building mechanical and electrical pipelines fail or are damaged, they will cause large property losses.

[0003] In the prior art, when conducting safety inspections on electromechanical pipelines, only the outer surface of the electromechanical pipelines and the temperature of the electromechanical pipelines themselves can be checked in a relatively simple manner, which results in the inability to detect safety hazards existing in the electromechanical pipelines. In addition, the prior art can only perform a single inspection on the electromechanical pipelines, which makes it impossible to monitor the conditions of the electromechanical pipelines in detail. Summary of the invention

[0004] Based on this, the purpose of the present invention is to provide an intelligent inspection method and system to at least solve the above-mentioned deficiencies in the prior art.

[0005] In a first aspect, the present invention provides an intelligent inspection method, the method comprising: Acquire the real-time status of the electromechanical pipeline based on several sensors on the support and hanger, and convert the real-time status into status text, wherein the status text includes vibration signal text, image signal text and temperature signal text; Using continuous wavelet transform and a preset number of decomposition layers to perform orthogonal decomposition on the vibration signal text under a preset admissibility condition to obtain a number of wavelet high-frequency coefficients, and processing the wavelet high-frequency coefficients based on an inverse wavelet transform to obtain a reconstructed denoised signal; The reconstructed denoised signal is transformed into a time-frequency diagram, the time-frequency domain characteristic parameters of the time-frequency diagram are calculated, and the time-frequency domain characteristic parameters are marked by using a threshold combined with a double-layer sliding window algorithm to obtain the disturbance of the pipeline; Segmenting the image signal text into a plurality of grid images, and predicting a class probability and a bounding box of each of the grid images, distinguishing predicted objects in the plurality of grid images based on the class probability and the bounding box, and filtering out bounding boxes of the predicted objects that are smaller than a preset confidence threshold based on non-maximum suppression to obtain a predicted image; Marking a suspected area in the predicted image, and acquiring a suspected target in the suspected area by mapping, so as to detect surface damage of the electromechanical pipeline; The image signal text and the temperature signal text are combined, and each temperature signal data in the temperature signal text is marked on a number of the grid images to obtain temperature information of the several grid images, and the temperature information of the several grid images is compared with the safety threshold temperature to obtain the temperature safety status of the electromechanical pipeline.

[0006] Compared with the prior art, the beneficial effects of the present invention are as follows: by adopting a threshold combined with a double-layer sliding window algorithm to mark the time-frequency domain feature parameters, the disturbance of the pipeline can be obtained, and then it can be judged whether the pipeline is operating normally. The predicted image obtained by filtering out the boundary box through non-maximum suppression, and the damage of the surface of the electromechanical pipeline can be monitored through calibration and mapping. By combining the image signal text with the temperature signal text, the temperature safety status of each area of ​​the electromechanical pipeline can be monitored.

[0007] Further, the step of converting the real-time status into status text, wherein the status text includes vibration signal text, image signal text and temperature signal text, comprises: Obtain vibration information, image information and temperature information of electromechanical pipelines; The vibration information, image information and temperature information are converted into vibration signal text, image signal text and temperature signal text based on a vector space model.

[0008] Furthermore, the step of processing a number of the wavelet high frequency coefficients based on the inverse wavelet transform to obtain a reconstructed denoised signal includes: Inputting a number of the wavelet high frequency coefficients into a preset threshold function for processing to obtain processed wavelet high frequency coefficients; The last layer of wavelet high-frequency coefficients among the plurality of wavelet high-frequency coefficients and the processed wavelet high-frequency coefficients are subjected to inverse wavelet transform to obtain a reconstructed denoised signal.

[0009] Furthermore, after the step of calculating the time-frequency domain characteristic parameters of the time-frequency graph, the method includes: Selecting a root mean square, a standard deviation, a peak factor, and a kurtosis in the time-frequency diagram, and using the root mean square, the standard deviation, the peak factor, and the kurtosis as degradation indicators; The degradation index is normalized, and the expression of the normalization is: ; In the formula, represents the original data features in the time-frequency diagram, express The maximum value of express The minimum value of It represents the normalized value of the degradation index.

[0010] Furthermore, the step of marking the time-frequency domain feature parameters by using a threshold combined with a double-layer sliding window algorithm includes: Preliminarily filtering the time-frequency domain feature parameters based on the small window, and performing a first mark on the filtered time-frequency domain feature parameters based on the large window to obtain a marked time-frequency domain feature, wherein the small window and the large window both include a number of points, and the number of points in the large window is greater than the number of points in the small window, and if more than half of the points in the small window exceed a set threshold, preliminary filtering is performed, and the first mark is performed through the large window; An initial fuzzy value is obtained, the initial fuzzy value is used as an initial threshold, and the initial threshold is adjusted based on the small window and the large window to perform a second marking.

[0011] Furthermore, the step of segmenting the image signal text into a plurality of grid images and predicting the class probability and bounding box of each of the grid images includes: Segmenting the image signal text into a plurality of grid images based on a YOLO network model; The class probability and bounding box of each of the grid images are predicted based on the YOLO network model according to the centers, heights, widths and classes to which the objects belong.

[0012] Furthermore, the step of calibrating the suspicious area in the predicted image and acquiring the suspicious target in the suspicious area by mapping includes: Using the Fast RCNN network structure model and based on the selective search algorithm, the suspicious area in the predicted image is calibrated; The suspected target in the suspected area is mapped based on the ROI pooling layer in the Fast RCNN network structure model.

[0013] Furthermore, the step of combining the image signal text and the temperature signal text and marking each temperature signal data in the temperature signal text on a plurality of the grid images includes: Classifying the image signal text by a binary mode to obtain an image classification set, and dividing the image classification set into a plurality of groups based on a decision plane to obtain a plurality of image grid sets of the same category; The temperature signal text is marked on a collection of several image grids of the same category to obtain the temperature conditions of each area of ​​the same category on the electromechanical pipeline.

[0014] Furthermore, the expression of the image classification set is: ; In the formula, represents the image dimension, , represents the image signal text, represents the data dimension, Indicates category, Indicates the total amount of input image text; The expression of the decision plane is: ; In the formula, Represents an image, represents the weight vector, represents the transposed weight vector, represents the sample bias, represents the data dimension, represents a collection of image grids of the same category, represents the weight vector for the collection of image grids of the same category, A grid of images representing all categories.

[0015] In a second aspect, the present invention further provides an intelligent inspection system, the system comprising: An acquisition conversion module is used to acquire the real-time status of the electromechanical pipeline based on a number of sensors on the support and hanger, and convert the real-time status into a status text, wherein the status text includes a vibration signal text, an image signal text, and a temperature signal text; A decomposition transformation module, used to perform orthogonal decomposition of the vibration signal text under a preset admissibility condition by using continuous wavelet transform and a preset number of decomposition layers to obtain a plurality of wavelet high-frequency coefficients, and to process the plurality of wavelet high-frequency coefficients based on an inverse wavelet transform to obtain a reconstructed denoised signal; A transformation calculation module, used for transforming the reconstructed denoised signal to obtain a time-frequency diagram, calculating the time-frequency domain characteristic parameters of the time-frequency diagram, and marking the time-frequency domain characteristic parameters by using a threshold combined with a double-layer sliding window algorithm to obtain the disturbance of the pipeline; A segmentation and filtering module, used to segment the image signal text into a plurality of grid images, and predict a class probability and a bounding box of each of the grid images, distinguish a plurality of predicted objects in the grid images based on the class probability and the bounding box, and filter out bounding boxes of the predicted objects that are smaller than a preset confidence threshold based on non-maximum suppression to obtain a predicted image; A calibration mapping module, used to calibrate the suspected area in the predicted image, and obtain the suspected target in the suspected area through mapping, so as to detect the surface damage of the electromechanical pipeline; A combined annotation module is used to combine the image signal text and the temperature signal text, and annotate each temperature signal data in the temperature signal text on a number of the grid images to obtain temperature information of the several grid images, and compare the temperature information of the several grid images with the safety threshold temperature to obtain the temperature safety status of the electromechanical pipeline. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 is a flow chart of the intelligent inspection method in the first embodiment of the present invention; Figure 2 It is a structural block diagram of the intelligent inspection system in the second embodiment of the present invention.

[0017] Description of main component symbols: 10. Get the conversion module; 20. Decomposition transformation module; 30. Transformation calculation module; 40. Segmentation filtering module; 50. Calibration mapping module; 60. Combined with annotation module.

[0018] The following specific implementation manner will further illustrate the present invention in conjunction with the above-mentioned drawings. DETAILED DESCRIPTION

[0019] In order to facilitate the understanding of the present invention, the present invention will be described more fully below with reference to the relevant drawings. Several embodiments of the present invention are given in the drawings. However, the present invention can be implemented in many different forms and is not limited to the embodiments described herein. On the contrary, the purpose of providing these embodiments is to make the disclosure of the present invention more thorough and comprehensive.

[0020] It should be noted that when an element is referred to as being "fixed to" another element, it may be directly on the other element or there may be a central element. When an element is considered to be "connected to" another element, it may be directly connected to the other element or there may be a central element at the same time. The terms "vertical", "horizontal", "left", "right" and similar expressions used herein are for illustrative purposes only.

[0021] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which the present invention belongs. The terms used herein in the specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. The term "and / or" used herein includes any and all combinations of one or more of the related listed items.

[0022] Embodiment 1 See also Figure 1 , which shows the intelligent inspection method in the first embodiment of the present invention, and the method includes steps S1 to S6: S1, obtaining the real-time status of the electromechanical pipeline based on a number of sensors on the support and hanger, and converting the real-time status into a status text, wherein the status text includes a vibration signal text, an image signal text, and a temperature signal text; Specifically, the step S1 includes steps S11 to S12: S11, obtaining vibration information, image information and temperature information of electromechanical pipelines; S12, converting the vibration information, image information and temperature information into vibration signal text, image signal text and temperature signal text based on a vector space model; It can be understood that the real-time status of the electromechanical pipeline is detected by several sensors on the support bracket, so as to obtain the vibration information, image information and temperature information in the electromechanical pipeline, and the features in the vibration information, image information and temperature information are extracted by the vector space model, so as to obtain the vibration signal text, image signal text and temperature signal text; It needs to be explained that the vector space model can describe the characteristic items of vibration information, image information and temperature information, so as to derive vibration signal text, image signal text and temperature signal text, and monitor the status of electromechanical pipelines based on the three angles of vibration signal text, image signal text and temperature signal text, thereby effectively improving the comprehensiveness of electromechanical pipeline status monitoring.

[0023] S2, using continuous wavelet transform and a preset number of decomposition layers to perform orthogonal decomposition on the vibration signal text under a preset admissibility condition to obtain a number of wavelet high-frequency coefficients, and processing the number of wavelet high-frequency coefficients based on an inverse wavelet transform to obtain a reconstructed denoised signal; It can be understood that the core of wavelet transform is the wavelet mother function. In this embodiment, the expression of the preset admissibility condition of the wavelet mother function is: ; In the formula, is the wavelet mother function Fourier transform, R represents the mean square integrable space, It represents the basic wavelet. Through the expression of the admissibility condition, it can be seen that the vibration signal has the characteristics of fluctuation and smallness. The two parameters in the continuous wavelet transform vibration signal text are translated and scaled, and then orthogonally decomposed, so as to obtain several wavelet high-frequency coefficients of the vibration signal text; It should be explained that, in this embodiment, the number of decomposition levels for decomposing the vibration signal text is five.

[0024] Specifically, the step S2 includes steps S21 to S22: S21, inputting a plurality of said wavelet high frequency coefficients into a preset threshold function for processing to obtain processed wavelet high frequency coefficients; S22, performing inverse wavelet transform on the last layer of wavelet high-frequency coefficients among the plurality of wavelet high-frequency coefficients and the processed wavelet high-frequency coefficients to obtain a reconstructed denoised signal; It is understandable that a reasonable threshold function is set, and then the wavelet high-frequency coefficients are input into the threshold function for processing. It is worth noting that in this embodiment, the wavelet high-frequency coefficients need to be input into the threshold function for processing in each layer of decomposition to obtain the wavelet high-frequency coefficients, which can effectively avoid the distortion of the vibration signal.

[0025] S3, transforming the reconstructed denoised signal to obtain a time-frequency diagram, calculating the time-frequency domain characteristic parameters of the time-frequency diagram, and marking the time-frequency domain characteristic parameters using a threshold combined with a double-layer sliding window algorithm to obtain the disturbance of the pipeline; It needs to be explained that the reconstructed denoised signal is converted by short-time Fourier transform, which mainly cuts the vibration signal according to each time period, obtains the frequency domain information of each time point by short-time Fourier transform, and then merges the signals of different time periods, and finally obtains the time-frequency diagram, and then calculates the variance and peak-to-peak value in the time-frequency diagram, and then uses the threshold combined with the double-layer sliding window algorithm to mark the time-frequency domain characteristic parameters, because when the pipeline is not disturbed, the time-frequency domain characteristic parameters are relatively smooth, and by marking the time-frequency domain characteristic parameters, some time-frequency domain characteristic parameters with large changes are marked, so that the disturbance of the pipeline can be obtained, that is, the abnormal operation of the pipeline can be monitored, and then the internal abnormal operation of the pipeline can be monitored; Specifically, the step S3 includes steps S31 to S32: S31, selecting a root mean square, a standard deviation, a peak factor, and a kurtosis in the time-frequency graph, and using the root mean square, the standard deviation, the peak factor, and the kurtosis as degradation indicators; S32, normalizing the degradation index, wherein the normalization expression is: ; In the formula, represents the original data features in the time-frequency diagram, express The maximum value of express The minimum value of represents the normalized value of the degradation index; It is worth mentioning that by taking the root mean square, standard deviation, peak factor and kurtosis of the features that can express the vibration signal text in the time-frequency diagram, and then using these features as degradation indicators, a new text can be obtained after processing the degradation indicators, which makes it possible to more accurately mark the subsequent marking process through the threshold combined with the double-layer sliding window algorithm.

[0026] In addition, in this embodiment, the step S3 further includes steps S33 to S34: S33, preliminarily filtering the time-frequency domain feature parameters based on the small window, and performing a first marking on the filtered time-frequency domain feature parameters based on the large window to obtain a marked time-frequency domain feature, wherein the small window and the large window both include a number of points, and the number of points in the large window is greater than the number of points in the small window, and if more than half of the points in the small window exceed a set threshold, preliminarily filtering is performed, and the first marking is performed through the large window; It can be understood that the double-layer sliding window algorithm includes two layers of windows. The first layer of windows is a small window. In this embodiment, the size of the small window is 25 points. If more than half of the 25 points in the small window exceed the set safety threshold, preliminary filtering is performed and then marked through the large window, which is the first mark, so that the marked time domain features can be obtained.

[0027] S34, obtaining an initial fuzzy value, taking the initial fuzzy value as an initial threshold, and adjusting the initial threshold based on the small window and the large window to perform a second marking; It can be understood that in this embodiment, there are 120 points in the large window. If more than half of the points in the large window exceed the safety threshold, a second mark is made on the points in the small window that exceed the set safety threshold, so that the abnormal situation of the pipeline can be obtained, and then the abnormal disturbance of the pipeline can be monitored.

[0028] It is worth mentioning that the vibration signals of the pipeline without disturbance are obtained and sorted according to the amplitude, the mean is proposed as the initial fuzzy value, and then the initial fuzzy value is finely adjusted to obtain the initial threshold, which can ensure the accuracy of the disturbance marking and maintain high sensitivity to disturbances with higher amplitudes.

[0029] S4, dividing the image signal text into a plurality of grid images, and predicting a class probability and a bounding box of each of the grid images, distinguishing a plurality of predicted objects in the grid images based on the class probability and the bounding box, and filtering out bounding boxes of the predicted objects that are smaller than a preset confidence threshold based on non-maximum suppression to obtain a predicted image; Specifically, the step S4 includes step S41 to step S42: S41, segmenting the image signal text into a plurality of grid images based on a YOLO network model; It can be understood that the YOLO network model applies the convolutional neural network to the entire image signal text, which can divide the entire image signal text into several grid images; S42, predicting a class probability and a bounding box of each of the grid images based on the YOLO network model according to the center, height, width and class of the object of the grid images; It can be understood that the YOLO network model in each grid image predicts a bounding box and the probability corresponding to each category, that is, the class probability. In this embodiment, each category includes pipes, lines, and protective sheaths. In this embodiment, the center of the grid image, the height of the grid image, the width of the grid image, and the class to which the object belongs are passed. If an object of one category falls into multiple grid images, then the grid image has multiple bounding boxes.

[0030] It needs to be explained that the predicted objects in the grid image are distinguished by class probability and bounding boxes, and then non-maximum suppression is used to filter out the bounding boxes of the predicted objects that are smaller than a preset confidence threshold in the grid image of each category object to obtain a predicted image. It can be understood that the predicted image is an image of the area that needs to be monitored, and non-maximum suppression, that is, suppressing elements that are not maximum values, can be understood as a local maximum search.

[0031] S5, calibrating a suspected area in the predicted image, and acquiring a suspected target in the suspected area by mapping, so as to detect surface damage of the electromechanical pipeline; Specifically, the step S5 includes step S51 to step S52: S51, using the Fast RCNN network structure model and based on a selective search algorithm to calibrate the suspicious area in the predicted image; S52, mapping the suspected target in the suspected area based on the ROI pooling layer in the Fast RCNN network structure model; It can be understood that after the predicted image is convolved in the ROI pooling layer in the Fast RCNN network structure model, the suspected area in the predicted image can be automatically calibrated through the selective search algorithm, and the suspected area is mapped through the ROI pooling layer in the FastRCNN network structure model, so that the suspected target can be obtained, and then the damage of the outer surface of the electromechanical pipeline can be monitored.

[0032] S6, combining the image signal text and the temperature signal text, and marking each temperature signal data in the temperature signal text on a plurality of the grid images to obtain temperature information of the plurality of the grid images, and comparing the temperature information of the plurality of the grid images with the safety threshold temperature to obtain the temperature safety status of the electromechanical pipeline; It can be understood that the image signal text and the temperature signal text are combined, and since the image signal text is divided into several grid images, the temperature of each area on the electromechanical pipeline can be marked through the combination of the temperature signal text and the image signal text, and then the temperature of each area is compared with the safety threshold temperature, so that the temperature safety status of each area of ​​the electromechanical pipeline can be obtained, thereby achieving the purpose of monitoring the electromechanical pipeline from multiple angles.

[0033] Specifically, the step S6 includes step S61 to step S62: S61, classifying the image signal text by a binary mode to obtain an image classification set, and dividing the image classification set into a plurality of groups based on a decision plane to obtain a plurality of image grid sets of the same category; S62, marking the temperature signal text on a collection of several image grids of the same category to obtain the temperature conditions of each area of ​​the same category on the electromechanical pipeline; The expression of the image classification set is: ; In the formula, represents the image dimension, , represents the image signal text, represents the data dimension, Indicates category, Indicates the total amount of input image text; The expression of the decision plane is: ; In the formula, Represents an image, represents the weight vector, represents the transposed weight vector, represents the sample bias, represents the data dimension, represents a collection of image grids of the same category, represents the weight vector for the collection of image grids of the same category, A grid of images representing all categories; It should be explained that images can be classified through binary patterns, for example, into electromechanical category 1 and pipeline category 1, and the electromechanical category images and pipeline category 1 images can be separated through the decision plane, thereby avoiding the problem of incorrect labeling when performing temperature labeling, because the operating temperatures of electromechanical category 1, pipeline category 1, and wire category 1 are different. Therefore, after separation, the temperature safety status of each area can be clearly judged, and the temperature safety status of the electromechanical pipeline can be effectively judged.

[0034] In summary, the intelligent inspection method in the above embodiment of the present invention can obtain the disturbance of the pipeline by marking the time-frequency domain feature parameters with a threshold combined with a double-layer sliding window algorithm, and then determine whether the pipeline is operating normally. The predicted image is obtained by filtering out the boundary box through non-maximum suppression, and the damage of the surface of the electromechanical pipeline can be monitored through calibration and mapping. By combining the image signal text with the temperature signal text, the temperature safety status of each area of ​​the electromechanical pipeline can be monitored.

[0035] Embodiment 2 See also Figure 2 , which shows an intelligent inspection system in a second embodiment of the present invention, the system comprises: An acquisition conversion module 10 is used to acquire the real-time status of the electromechanical pipeline based on a number of sensors on the support and hanger, and convert the real-time status into a status text, wherein the status text includes a vibration signal text, an image signal text, and a temperature signal text; The decomposition transformation module 20 is used to perform orthogonal decomposition on the vibration signal text under a preset admissibility condition by using continuous wavelet transform and a preset decomposition layer number to obtain a plurality of wavelet high-frequency coefficients, and to process the plurality of wavelet high-frequency coefficients based on an inverse wavelet transform to obtain a reconstructed denoised signal; A transformation calculation module 30 is used to transform the reconstructed denoised signal to obtain a time-frequency diagram, calculate the time-frequency domain characteristic parameters of the time-frequency diagram, and mark the time-frequency domain characteristic parameters using a threshold combined with a double-layer sliding window algorithm to obtain the disturbance of the pipeline; A segmentation and filtering module 40 is used to segment the image signal text into a plurality of grid images, and predict a class probability and a bounding box of each of the grid images, distinguish a plurality of predicted objects in the grid images based on the class probability and the bounding box, and filter out bounding boxes of the predicted objects that are smaller than a preset confidence threshold based on non-maximum suppression to obtain a predicted image; A calibration mapping module 50, used to calibrate the suspected area in the predicted image, and obtain the suspected target in the suspected area by mapping, so as to detect the surface damage of the electromechanical pipeline; Combined with the annotation module 60, it is used to combine the image signal text and the temperature signal text, and annotate each temperature signal data in the temperature signal text on a number of the grid images to obtain the temperature information of the several grid images, and compare the temperature information of the several grid images with the safety threshold temperature to obtain the temperature safety status of the electromechanical pipeline.

[0036] In some optional embodiments, the acquisition conversion module 10 includes: An acquisition unit, used to acquire vibration information, image information and temperature information of electromechanical pipelines; The conversion unit is used to convert the vibration information, image information and temperature information into vibration signal text, image signal text and temperature signal text based on a vector space model.

[0037] In some optional embodiments, the decomposition transformation module 20 includes: A processing unit, used for inputting a plurality of said wavelet high frequency coefficients into a preset threshold function for processing to obtain processed wavelet high frequency coefficients; The transformation processing unit is used to perform inverse wavelet transformation on the last layer of wavelet high-frequency coefficients among the plurality of wavelet high-frequency coefficients and the processed wavelet high-frequency coefficients to obtain a reconstructed denoised signal.

[0038] In some optional embodiments, the transformation calculation module 30 includes: A selection unit, used for selecting a root mean square, a standard deviation, a peak factor and a kurtosis in the time-frequency diagram, and using the root mean square, the standard deviation, the peak factor and the kurtosis as degradation indicators; A normalization processing unit is used to perform normalization processing on the degradation index, and the expression of the normalization processing is: ; In the formula, represents the original data features in the time-frequency diagram, express The maximum value of express The minimum value of represents the normalized value of the degradation index; A first marking unit is used to perform preliminary filtering on the time-frequency domain feature parameters based on a small window, and perform a first marking on the filtered time-frequency domain feature parameters based on a large window to obtain a marked time-frequency domain feature, wherein the small window and the large window both include a number of points, and the number of points in the large window is greater than the number of points in the small window, and if more than half of the points in the small window exceed a set threshold, preliminary filtering is performed, and the first marking is performed through the large window; The second marking unit is used to obtain an initial fuzzy value, use the initial fuzzy value as an initial threshold, and adjust the initial threshold based on the small window and the large window to perform a second marking.

[0039] In some optional embodiments, the segmentation and filtering module 40 includes: A segmentation unit, used for segmenting the image signal text into a plurality of grid images based on a YOLO network model; A prediction unit is used to predict the class probability and bounding box of each of the grid images based on the YOLO network model according to the center, height, width and class of the object of the grid images.

[0040] In some optional embodiments, the calibration mapping module 50 includes: A calibration unit, used for calibrating the suspicious area in the predicted image using a Fast RCNN network structure model and based on a selective search algorithm; A mapping unit is used to map the suspected target in the suspected area based on the ROI pooling layer in the Fast RCNN network structure model.

[0041] In some optional embodiments, the combined annotation module 60 includes: A classification unit, used for classifying the image signal text by a binary mode to obtain an image classification set, and dividing the image classification set into a plurality of groups based on a decision plane to obtain a plurality of image grid sets of the same category; A labeling unit, used for labeling the temperature signal text on a collection of several image grids of the same category, so as to obtain the temperature conditions of each area of ​​the same category on the electromechanical pipeline; The expression of the image classification set is: ; In the formula, represents the image dimension, , represents the image signal text, represents the data dimension, Indicates category, Indicates the total amount of input image text; The expression of the decision plane is: ; In the formula, Represents an image, represents the weight vector, represents the transposed weight vector, represents the sample bias, represents the data dimension, represents a collection of image grids of the same category, represents the weight vector for the collection of image grids of the same category, A grid of images representing all categories.

[0042] The functions or operation steps implemented when the above modules and units are executed are generally the same as those in the above method embodiments, and will not be repeated here.

[0043] The intelligent inspection system provided in the embodiment of the present invention has the same implementation principle and technical effects as those of the aforementioned method embodiment. For the sake of brief description, for matters not mentioned in the system embodiment, reference may be made to the corresponding contents in the aforementioned method embodiment.

[0044] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "examples", "specific examples", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.

[0045] The above-mentioned embodiments only express several implementation methods of the present invention, and the description thereof is relatively specific and detailed, but it cannot be understood as limiting the scope of the patent of the present invention. It should be pointed out that, for ordinary technicians in this field, several variations and improvements can be made without departing from the concept of the present invention, which all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be subject to the attached claims.

Claims

1. An intelligent inspection method, characterized in that: The method comprises: Acquire the real-time status of the electromechanical pipeline based on several sensors on the support and hanger, and convert the real-time status into status text, wherein the status text includes vibration signal text, image signal text and temperature signal text; Using continuous wavelet transform and a preset number of decomposition layers to perform orthogonal decomposition on the vibration signal text under a preset admissibility condition to obtain a number of wavelet high-frequency coefficients, and processing the wavelet high-frequency coefficients based on an inverse wavelet transform to obtain a reconstructed denoised signal; The reconstructed denoised signal is transformed into a time-frequency diagram, the time-frequency domain characteristic parameters of the time-frequency diagram are calculated, and the time-frequency domain characteristic parameters are marked by using a threshold combined with a double-layer sliding window algorithm to obtain the disturbance of the pipeline; Segmenting the image signal text into a plurality of grid images, and predicting a class probability and a bounding box of each of the grid images, distinguishing predicted objects in the plurality of grid images based on the class probability and the bounding box, and filtering out bounding boxes of the predicted objects that are smaller than a preset confidence threshold based on non-maximum suppression to obtain a predicted image; Marking a suspected area in the predicted image, and acquiring a suspected target in the suspected area by mapping, so as to detect surface damage of the electromechanical pipeline; The image signal text and the temperature signal text are combined, and each temperature signal data in the temperature signal text is marked on a number of the grid images to obtain temperature information of the several grid images, and the temperature information of the several grid images is compared with the safety threshold temperature to obtain the temperature safety status of the electromechanical pipeline.

2. The intelligent inspection method according to claim 1, characterized in that: The step of converting the real-time status into a status text, wherein the status text includes a vibration signal text, an image signal text, and a temperature signal text comprises: Obtain vibration information, image information and temperature information of electromechanical pipelines; The vibration information, image information and temperature information are converted into vibration signal text, image signal text and temperature signal text based on a vector space model.

3. The intelligent inspection method according to claim 1, characterized in that: The step of processing a number of the wavelet high frequency coefficients based on the inverse wavelet transform to obtain a reconstructed denoised signal comprises: Inputting a number of the wavelet high frequency coefficients into a preset threshold function for processing to obtain processed wavelet high frequency coefficients; The last layer of wavelet high-frequency coefficients among the plurality of wavelet high-frequency coefficients and the processed wavelet high-frequency coefficients are subjected to inverse wavelet transform to obtain a reconstructed denoised signal.

4. The intelligent inspection method according to claim 1, characterized in that: After the step of calculating the time-frequency domain characteristic parameters of the time-frequency graph, the method comprises: Selecting a root mean square, a standard deviation, a peak factor, and a kurtosis in the time-frequency diagram, and using the root mean square, the standard deviation, the peak factor, and the kurtosis as degradation indicators; The degradation index is normalized, and the expression of the normalization is: ; In the formula, represents the original data features in the time-frequency diagram, express The maximum value of express The minimum value of It represents the normalized value of the degradation index.

5. The intelligent inspection method according to claim 1, characterized in that: The step of using a threshold combined with a double-layer sliding window algorithm to mark the time-frequency domain feature parameters comprises: Preliminarily filtering the time-frequency domain feature parameters based on the small window, and performing a first mark on the filtered time-frequency domain feature parameters based on the large window to obtain a marked time-frequency domain feature, wherein the small window and the large window both include a number of points, and the number of points in the large window is greater than the number of points in the small window, and if more than half of the points in the small window exceed a set threshold, preliminary filtering is performed, and the first mark is performed through the large window; An initial fuzzy value is obtained, the initial fuzzy value is used as an initial threshold, and the initial threshold is adjusted based on the small window and the large window to perform a second marking.

6. The intelligent inspection method according to claim 1, characterized in that: The step of segmenting the image signal text into a plurality of grid images and predicting the class probability and bounding box of each of the grid images comprises: Segmenting the image signal text into a plurality of grid images based on a YOLO network model; According to the center, height, width and class of the object of the plurality of grid images and based on the YOLO network model, the class probability and bounding box of each grid image are predicted.

7. The intelligent inspection method according to claim 1, characterized in that: The step of calibrating the suspicious area in the predicted image and acquiring the suspicious target in the suspicious area by mapping comprises: Using the Fast RCNN network structure model and based on the selective search algorithm, the suspicious area in the predicted image is calibrated; The suspected target in the suspected area is mapped based on the ROI pooling layer in the Fast RCNN network structure model.

8. The intelligent inspection method according to claim 1, characterized in that: The step of combining the image signal text and the temperature signal text and marking each temperature signal data in the temperature signal text on a plurality of the grid images comprises: Classifying the image signal text by a binary mode to obtain an image classification set, and dividing the image classification set into a plurality of groups based on a decision plane to obtain a plurality of image grid sets of the same category; The temperature signal text is marked on a collection of several image grids of the same category to obtain the temperature conditions of each area of ​​the same category on the electromechanical pipeline.

9. The intelligent inspection method according to claim 8, characterized in that: The expression of the image classification set is: ; In the formula, represents the image dimension, , represents the image signal text, represents the data dimension, Indicates category, Indicates the total amount of input image text; The expression of the decision plane is: ; In the formula, Represents an image, represents the weight vector, represents the transposed weight vector, represents the sample bias, represents the data dimension, represents a collection of image grids of the same category, represents the weight vector for the collection of image grids of the same category, A grid of images representing all categories.

10. An intelligent inspection system, characterized in that: The system comprises: An acquisition conversion module is used to acquire the real-time status of the electromechanical pipeline based on a number of sensors on the support and hanger, and convert the real-time status into a status text, wherein the status text includes a vibration signal text, an image signal text, and a temperature signal text; A decomposition transformation module, used to perform orthogonal decomposition of the vibration signal text under a preset admissibility condition by using continuous wavelet transform and a preset number of decomposition layers to obtain a plurality of wavelet high-frequency coefficients, and to process the plurality of wavelet high-frequency coefficients based on an inverse wavelet transform to obtain a reconstructed denoised signal; A transformation calculation module, used for transforming the reconstructed denoised signal to obtain a time-frequency diagram, calculating the time-frequency domain characteristic parameters of the time-frequency diagram, and marking the time-frequency domain characteristic parameters by using a threshold combined with a double-layer sliding window algorithm to obtain the disturbance of the pipeline; A segmentation and filtering module, used to segment the image signal text into a plurality of grid images, and predict a class probability and a bounding box of each of the grid images, distinguish a plurality of predicted objects in the grid images based on the class probability and the bounding box, and filter out bounding boxes of the predicted objects that are smaller than a preset confidence threshold based on non-maximum suppression to obtain a predicted image; A calibration mapping module, used to calibrate the suspected area in the predicted image, and obtain the suspected target in the suspected area through mapping, so as to detect the surface damage of the electromechanical pipeline; A combined annotation module is used to combine the image signal text and the temperature signal text, and annotate each temperature signal data in the temperature signal text on a number of the grid images to obtain temperature information of the several grid images, and compare the temperature information of the several grid images with the safety threshold temperature to obtain the temperature safety status of the electromechanical pipeline.

Citation Information

Patent Citations

  • Vibration acceleration signal denoising method and device based on wavelet transform

    CN114186577A

  • Power equipment operation state detection method based on infrared image

    CN114694050A

  • Detection method and system for multiple target images

    CN114927236A

  • Subway tunnel inspection system and inspection method

    CN115600124A

  • Remote inspection system for engineering construction site

    CN118673462A