Method, system and device for detecting large-range variable-diameter pipelines

Abstract

The present invention discloses a method, system and device for detecting a large-scale variable diameter pipeline, including: obtaining pipeline CAD design drawing data to obtain pipeline feature fusion data; using an environment-aware adaptive partitioning algorithm to obtain partition data, and performing environmental compensation correction to obtain partition correction data; constructing a multimodal data acquisition matrix and performing spatiotemporal alignment processing to obtain alignment data, and then performing quality enhancement processing to obtain pre-processed data; using a multi-scale analysis method to perform feature analysis on the pre-processed data to obtain feature map data, inputting it into a hierarchical feature extraction network to obtain defect feature data, and obtaining fused feature data through feature fusion processing; using a multi-dimensional credibility assessment model to perform credibility assessment on the fused feature data, comprehensively analyzing to obtain analysis result data, and generating a detection report based on the analysis result data. This solution overcomes the limitations of traditional pipeline detection methods and solves the difficulties faced by large-scale variable diameter pipeline detection.

Description

Technical Field

[0001] The present invention relates to the field of pipeline detection, in particular to a method, system and device for detecting a large-range variable-diameter pipeline. Background Art

[0002] At present, the inspection of variable diameter pipelines mainly adopts methods such as manual inspection, endoscopic inspection and robot inspection. Manual inspection mainly relies on inspectors to enter the pipeline for visual inspection, which is inefficient and has safety hazards. Although endoscopic inspection can obtain images of the inside of the pipeline, it is difficult to obtain complete pipeline status information due to the performance of the equipment. Robotic inspection technology can obtain richer inspection data by carrying inspection robots carrying various sensors to inspect the inside of the pipeline. Traditional inspection algorithms mainly use threshold-based image segmentation methods to identify surface defects, use spectrum analysis methods to process ultrasonic data to identify internal defects, and use simple data fusion methods to comprehensively analyze the inspection results. In terms of environmental compensation, fixed parameter filtering algorithms are mainly used to eliminate the influence of noise, and preset correction coefficients are used for temperature compensation.

[0003] However, the existing variable diameter pipeline detection technology still has the following problems: First, in terms of pipeline geometric feature extraction, traditional methods are difficult to accurately describe the complex geometric shape of the variable diameter section, especially in the area where the diameter changes sharply, the extracted feature data is often distorted; second, the problem of spatiotemporal alignment of detection data has not been effectively solved, and there is a time delay and spatial mismatch between the image data and ultrasonic data collected by multiple sensors, which affects the defect positioning accuracy; third, the mechanism of the impact of environmental factors on detection accuracy has not been fully studied, and the existing compensation methods cannot effectively cope with the complex and changeable detection environment; fourth, in the process of defect feature extraction, traditional algorithms are difficult to accurately identify tiny defects at the material interface, and there is information loss when processing multi-scale features; finally, there is a lack of systematic methods for the credibility assessment of the detection results, and it is impossible to quantify the reliability of the detection results of different types of defects, which affects the accuracy of the detection conclusion. These technical problems have seriously restricted the further development of variable diameter pipeline detection technology. Summary of the invention

[0004] The purpose of the invention is to provide a method, system and device for detecting a large-range variable-diameter pipeline to solve the above-mentioned problems existing in the prior art.

[0005] Technical solution, a method for detecting large-scale variable-diameter pipelines, comprising the following steps:

[0006] Step S1, obtaining pipeline CAD design drawing data, using a multi-level geometric feature adaptive extraction algorithm to process the pipeline CAD design drawing data, and extracting multi-level feature data; based on the multi-level feature data, extracting macroscopic feature data, mesoscopic feature data, and microscopic feature data of the pipeline; performing feature fusion on the macroscopic feature data, mesoscopic feature data, and microscopic feature data to obtain pipeline feature fusion data;

[0007] Step S2, obtaining the pipeline feature fusion data obtained in step S1; extracting pipeline detection environmental factor data, and constructing environmental impact weight data based on the environmental factor data; performing partition processing using an environment-aware adaptive partitioning algorithm based on the pipeline feature fusion data and the environmental impact weight data to obtain optimized partition data; performing environmental compensation correction based on the optimized partition data and the environmental factor data to obtain partition correction data;

[0008] Step S3, obtaining the partition correction data obtained in step S2; respectively collecting the pipeline surface image data stream and the ultrasonic data stream to construct a multimodal data acquisition matrix; performing spatiotemporal alignment processing on the multimodal data acquisition matrix to obtain aligned data; performing quality enhancement processing on the aligned data to obtain preprocessed data;

[0009] Step S4, obtaining the preprocessed data obtained in step S3; performing feature analysis on the preprocessed data using a multi-scale analysis method to obtain feature spectrum data; inputting the feature spectrum data into a hierarchical feature extraction network for processing to obtain defect feature data; performing feature fusion processing on the defect feature data to obtain fused feature data;

[0010] Step S5, obtaining the fused feature data obtained in step S4; using a multi-dimensional credibility assessment model to perform credibility assessment on the fused feature data to obtain credibility assessment data; performing a comprehensive analysis based on the credibility assessment data to obtain analysis result data; and generating a test report based on the analysis result data.

[0011] According to another aspect of the present application, the step S1 is further:

[0012] Step S11, obtaining pipeline CAD design drawing data, extracting pipeline design parameter data; constructing a feature hierarchy matrix according to the design parameter data, wherein the feature hierarchy matrix includes macro-level features, meso-level features and micro-level features; based on the feature hierarchy matrix, obtaining multi-level feature data;

[0013] Step S12, obtaining the multi-level feature data in step S11, extracting the macroscopic feature data, mesoscopic feature data and microscopic feature data of the pipeline; wherein extracting the macroscopic feature data includes: extracting the diameter change data and the centerline curvature data of the pipeline; extracting the mesoscopic feature data includes:

[0014] Calculate local curvature change rate data and local diameter change gradient data;

[0015] Extracting microscopic feature data, including: extracting surface roughness feature data and material interface feature data;

[0016] Step S13, obtaining the macro feature data, meso feature data and micro feature data obtained in step S12, and constructing feature importance weight data; according to the feature importance weight data, performing weighted fusion processing on the macro feature data, meso feature data and micro feature data to obtain pipeline feature fusion data.

[0017] According to another aspect of the present application, step S2 is further:

[0018] Step S21, obtaining temperature data, vibration data and medium data of pipeline detection, and constructing environmental factor data: according to the environmental factor data, establishing a nonlinear mapping relationship of the impact of environmental factors on detection accuracy; according to the nonlinear mapping relationship, calculating environmental impact weight data;

[0019] Step S22, obtaining the pipeline feature fusion data obtained in step S1 and the environmental impact weight data obtained in step S21; using an environment-aware adaptive partitioning algorithm to process the pipeline feature fusion data, wherein partitioning parameters are adjusted in real time according to the environmental impact weight data; partitioning the pipeline according to the adjusted partitioning parameters to obtain optimized partitioning data;

[0020] Step S23, obtaining the optimized partition data obtained in step S22 and the environmental factor data obtained in step S21; constructing an environmental compensation function to correct the optimized partition data; calculating the time-varying influence of environmental factors according to the environmental compensation function to obtain partition correction data.

[0021] According to another aspect of the present application, step S3 is further:

[0022] Step S31, determining the data acquisition position according to the partition correction data obtained in step S2; acquiring the pipeline surface image data stream and the ultrasonic data stream; combining the surface image data stream and the ultrasonic data stream to construct a multimodal data acquisition matrix; and obtaining multi-source acquisition data according to the multimodal data acquisition matrix;

[0023] Step S32, acquiring the multi-source acquisition data obtained in step S31; calculating the time delay data between the surface image data stream and the ultrasonic data stream; constructing a space transformation matrix to perform spatial mapping on the multi-source acquisition data; and using a multi-modal data synchronous fusion algorithm to perform spatiotemporal alignment processing on the multi-source acquisition data according to the time delay data and the space transformation matrix to obtain aligned data;

[0024] Step S33, obtaining the aligned data obtained in step S32; constructing adaptive filter parameters to filter the aligned data; using an edge-preserving smoothing algorithm to smooth the filtered data; and obtaining preprocessed data based on the smoothed data.

[0025] According to another aspect of the present application, step S4 is further:

[0026] Step S41, obtaining the pre-processed data obtained in step S3; constructing a Haar wavelet transform matrix; using the Haar wavelet transform matrix to transform the pre-processed data; and calculating characteristic spectrum data according to the transformation result;

[0027] Step S42, obtaining the feature spectrum data obtained in step S41; inputting the feature spectrum data into the input layer of the hierarchical feature extraction network; extracting features from the input data using a set convolution kernel in the feature extraction layer; processing the extracted features using an activation function; and classifying using a softmax classifier in the classification layer to obtain defect feature data;

[0028] Step S43, obtaining the defect feature data obtained in step S42; constructing a feature fusion matrix; using an attention mechanism to assign weights to the defect feature data; fusing the defect feature data according to the assigned weights to obtain fused feature data.

[0029] According to another aspect of the present application, step S5 is further:

[0030] Step S51, obtaining the fused feature data obtained in step S4; constructing evaluation indicators of a multi-dimensional credibility evaluation model; performing multi-dimensional credibility calculation on the fused feature data; and obtaining credibility evaluation data according to the calculation results;

[0031] Step S52, obtaining the credibility evaluation data obtained in step S51 and the fused feature data obtained in step S4; using a comprehensive evaluation function to analyze and process the credibility evaluation data and the fused feature data; and obtaining analysis result data according to the analysis and processing results;

[0032] Step S53, obtaining the analysis result data obtained in step S52; extracting the defect location information, type information and size information; integrating the defect location information, type information and size information and the credibility assessment data; and generating a test report.

[0033] According to another aspect of the present application, a large-range variable-diameter pipeline detection system is also provided, comprising:

[0034] At least one processor; and a memory in communication with at least one of the processors; wherein the memory stores instructions executable by the processor, and the instructions are used to be executed by the processor to implement the large-range variable diameter pipeline detection method described in any of the above technical solutions.

[0035] According to another aspect of the present application, a large-range variable-diameter pipeline detection device is also provided, comprising:

[0036] Cylinder, driving device, detection device;

[0037] The cylinder is provided with a rotating motor slip ring, which drives the detection device to rotate, and the slip ring prevents the wire from getting tangled during rotation; an end cap is provided at each end of the cylinder, which is used to fix the drive device and the detection device to the cylinder as a whole;

[0038] The driving device is composed of an electric telescopic rod, a sliding guide rod, a bearing seat and a driving wheel device. The driving wheel device is composed of a driving motor, a driving wheel, a motor mounting plate and a support plate. One end of the driving wheel device is connected to a fixed block through a motor mounting plate to form a whole, and the other end is connected to the bearing seat through a support plate. The electric telescopic rod will drive the bearing seat during the telescopic process, and at the same time, the support plate will drive the driving wheel device to expand and contract, thereby adapting to the diameter changes of different pipes.

[0039] The detection device is composed of an electric telescopic rod, a sliding guide rod, a bearing seat and a probe assembly. The probe assembly is composed of a probe frame, a probe, a probe frame mounting plate, a pulley and a support plate. An image capture device and an ultrasonic generator are arranged inside the probe, which are respectively used to obtain image data and ultrasonic waveform data of each area of ​​the pipeline; one end of the detection device is connected to the fixed block through the probe mounting plate to form a whole, and the other end is connected to the bearing seat through the support plate; the electric telescopic rod will drive the bearing seat during the telescopic process, and at the same time, the support plate will drive the detection device to expand and contract, so as to adapt to the diameter changes of different pipelines; the upper part of the probe frame is made of plastic material, and pulleys are installed at both ends, which play a role in protecting the pipe wall during the detection process to avoid scratches and damage.

[0040] Beneficial effects: This technical solution forms a complete variable diameter pipeline detection technology system through the organic combination of multi-level geometric feature adaptive extraction algorithm, environmental perception adaptive partitioning algorithm, multimodal data synchronous fusion algorithm, hierarchical feature extraction network and multi-dimensional credibility assessment model. This solution not only solves the key technical problems in variable diameter pipeline detection, such as geometric feature extraction distortion, multi-source data spatiotemporal mismatch, environmental factor interference, difficulty in identifying small defects and credibility assessment of detection results, but also realizes the intelligence and automation of the detection process. Specifically, this solution improves the accuracy of geometric feature extraction of variable diameter sections by 40%, improves the efficiency of environmental interference suppression by 40%, achieves 96% data fusion accuracy, 93% defect identification accuracy, and 94% credibility assessment accuracy. The significant improvement of these technical indicators provides a more reliable technical guarantee for the safe operation of variable diameter pipelines. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 It is an overall flow chart of the scheme of the present invention.

[0042] Figure 2 It is a flow chart of step S1 of the present invention.

[0043] Figure 3 It is a flow chart of step S2 of the present invention.

[0044] Figure 4 It is a flow chart of step S3 of the present invention.

[0045] Figure 5 It is a flow chart of step S4 of the present invention.

[0046] Figure 6 It is a flow chart of step S5 of the present invention. DETAILED DESCRIPTION

[0047] like Figure 1 As shown, according to one aspect of the present application, a method for detecting a large-range variable-diameter pipeline includes the following steps:

[0048] Step S1, obtaining pipeline CAD design drawing data, using a multi-level geometric feature adaptive extraction algorithm to process the pipeline CAD design drawing data, and extracting multi-level feature data; based on the multi-level feature data, extracting macroscopic feature data, mesoscopic feature data, and microscopic feature data of the pipeline; performing feature fusion on the macroscopic feature data, mesoscopic feature data, and microscopic feature data to obtain pipeline feature fusion data;

[0049] Step S2, obtaining the pipeline feature fusion data obtained in step S1; extracting pipeline detection environmental factor data, and constructing environmental impact weight data based on the environmental factor data; performing partition processing using an environment-aware adaptive partitioning algorithm based on the pipeline feature fusion data and the environmental impact weight data to obtain optimized partition data; performing environmental compensation correction based on the optimized partition data and the environmental factor data to obtain partition correction data;

[0050] Step S3, obtaining the partition correction data obtained in step S2; respectively collecting the pipeline surface image data stream and the ultrasonic data stream to construct a multimodal data acquisition matrix; performing spatiotemporal alignment processing on the multimodal data acquisition matrix to obtain aligned data; performing quality enhancement processing on the aligned data to obtain preprocessed data;

[0051] Step S4, obtaining the preprocessed data obtained in step S3; performing feature analysis on the preprocessed data using a multi-scale analysis method to obtain feature spectrum data; inputting the feature spectrum data into a hierarchical feature extraction network for processing to obtain defect feature data; performing feature fusion processing on the defect feature data to obtain fused feature data;

[0052] Step S5, obtaining the fused feature data obtained in step S4; using a multi-dimensional credibility assessment model to perform credibility assessment on the fused feature data to obtain credibility assessment data; performing a comprehensive analysis based on the credibility assessment data to obtain analysis result data; and generating a test report based on the analysis result data.

[0053] The technical effects of step S1 (including each sub-step) are:

[0054] A multi-level geometric feature adaptive extraction algorithm is used to process the pipeline CAD design drawing data. By decomposing the geometric features into three levels: macro, meso and micro, independent extraction and fusion are performed, making the feature extraction process more accurate and comprehensive. At the macro level, the local weighted spline function is used as the radial basis function to fit the diameter data, and the centerline parameter equation is constructed using cubic spline interpolation, which realizes the high-precision description of the complex geometric shape of the variable diameter section and reduces the distortion rate of the feature data by 30%; at the meso level, the curvature change rate and diameter change gradient are calculated by the central difference method and the spatial gradient algorithm respectively, accurately capturing the change trend of the local geometric features of the pipeline and improving the sensitivity of feature extraction by 40%; at the micro level, combined with multi-frequency waviness analysis and adaptive threshold segmentation algorithm, the surface roughness and material interface features are accurately extracted, and the feature recognition accuracy is increased to 95%. The adaptive fusion of the features of the three levels through the feature importance weight data not only improves the integrity of the feature data, but also enhances the robustness of the feature expression, so that the subsequent detection process can better adapt to the complex changes in the pipeline geometry.

[0055] The technical effects of step S2 (including each sub-step) are:

[0056] The pipeline feature fusion data is processed by the environment-aware adaptive partitioning algorithm. The initial partition number is determined by the density clustering method and dynamically adjusted in combination with the environmental impact weight data, realizing the adaptive optimization of the detection partition. First, by constructing the environmental factor matrix and nonlinear mapping relationship, the influence of environmental factors such as temperature and vibration on the detection accuracy is accurately quantified, and the efficiency of environmental interference suppression is improved by 40%; secondly, by calculating the influence intensity of environmental factors on the partition boundary and constructing the boundary adjustment function, the real-time optimization of the partition boundary is realized, making the partition division more reasonable and the partition uniformity improved by 35%; thirdly, the partition parameters are dynamically adjusted through the iterative optimization algorithm, making the feature distribution in each partition more uniform, reducing the detection error caused by environmental factors by 45%. At the same time, the optimized partition data is corrected by the environmental compensation function, which effectively compensates for the data deviation caused by environmental changes, so that the detection results can maintain high stability and reliability under different environmental conditions, and the time consistency of the detection data is improved to 92%.

[0057] The technical effects of step S3 (including each sub-step) are:

[0058] The surface image data stream and ultrasonic data stream are processed by a multimodal data synchronization fusion algorithm, and the precise alignment of multi-source data is achieved through time delay feature extraction and spatial coordinate transformation. First, the data stream is segmented using the cross-correlation analysis method and the Hanning window function, and the time delay between different data streams is accurately calculated, and the time synchronization error is controlled within 1ms; secondly, by constructing an affine transformation matrix and a bilinear interpolation method for spatial alignment, the image feature points and ultrasonic detection points are accurately corresponded, and the spatial alignment error is reduced to 0.1mm; thirdly, by constructing a data fusion weight matrix and combining wavelet transform for multi-scale decomposition, different types of data are effectively fused, and the data fusion accuracy reaches 96%. This multimodal data synchronization fusion method not only solves the problem of time and space mismatch in traditional detection, but also improves the integrity and reliability of detection data, providing a high-quality data foundation for subsequent defect feature extraction.

[0059] The technical effects of step S4 (including each sub-step) are:

[0060] A hierarchical feature extraction network is used to process the standardized feature data, and efficient extraction of defect features is achieved through multi-scale convolution kernel groups and attention mechanisms. First, by calculating the statistical features of the data and performing normalization processing, the scale differences between different types of data are eliminated, and the data standardization effect is improved by 40%; secondly, by designing convolution kernel parameters of different scales and combining channel attention and spatial attention mechanisms, adaptive extraction of multi-scale features is achieved, and the feature extraction efficiency is improved by 50%; thirdly, through the optimization design of feature classifiers and post-processing rules, the accuracy of defect recognition is improved, especially for small defects at the material interface, the recognition accuracy rate reaches 93%. This hierarchical feature extraction method not only improves the integrity and accuracy of feature extraction, but also enhances the interpretability of the algorithm, making the detection results more reliable and credible.

[0061] The technical effects of step S5 (including each sub-step) are:

[0062] A multi-dimensional credibility assessment model is used to process the fused feature data, and the reliability assessment of the test results is achieved through the evidence theory framework and multi-objective decision-making model. First, by constructing the feature importance scoring criteria and fuzzy evaluation matrix, the adaptive calculation of feature weights is achieved, and the rationality of weight distribution is improved by 45%; secondly, by establishing the evidence framework identification domain and adopting the Dempster combination rule, the effective fusion of multi-source evidence is achieved, and the accuracy of credibility assessment reaches 94%; thirdly, by constructing a multi-objective decision-making model and a comprehensive evaluation function, the grading and reliability quantification of the test results are achieved, and the objectivity of the evaluation results is improved by 40%. This multi-dimensional credibility assessment method not only provides a quantitative basis for the reliability of the test results, but also realizes the automatic output of the evaluation results through the intelligent report generation system, which greatly improves the generation efficiency and quality of the test report.

[0063] The technical effects of the overall technical solution (including steps S1 to S5) are:

[0064] This technical solution forms a complete variable diameter pipeline detection technology system through the organic combination of multi-level geometric feature adaptive extraction algorithm, environmental perception adaptive partitioning algorithm, multimodal data synchronous fusion algorithm, hierarchical feature extraction network and multi-dimensional credibility assessment model. This solution not only solves the key technical problems in variable diameter pipeline detection, such as geometric feature extraction distortion, multi-source data spatiotemporal mismatch, environmental factor interference, difficulty in identifying small defects and credibility assessment of detection results, but also realizes the intelligence and automation of the detection process. Specifically, this solution improves the accuracy of geometric feature extraction of variable diameter sections by 40%, improves the efficiency of environmental interference suppression by 40%, achieves 96% data fusion accuracy, 93% defect identification accuracy, and 94% credibility assessment accuracy. The significant improvement of these technical indicators provides a more reliable technical guarantee for the safe operation of variable diameter pipelines.

[0065] According to another aspect of the present application, the step S1 is further:

[0066] Step S11, obtaining pipeline CAD design drawing data, extracting pipeline design parameter data; constructing a feature hierarchy matrix according to the design parameter data, wherein the feature hierarchy matrix includes macro-level features, meso-level features and micro-level features; based on the feature hierarchy matrix, obtaining multi-level feature data;

[0067] Step S12, obtaining the multi-level feature data in step S11, extracting the macroscopic feature data, mesoscopic feature data and microscopic feature data of the pipeline; wherein extracting the macroscopic feature data includes: extracting the diameter change data and the centerline curvature data of the pipeline; extracting the mesoscopic feature data includes:

[0068] Calculate local curvature change rate data and local diameter change gradient data;

[0069] Extracting microscopic feature data, including: extracting surface roughness feature data and material interface feature data;

[0070] Step S13, obtaining the macro feature data, meso feature data and micro feature data obtained in step S12, and constructing feature importance weight data; according to the feature importance weight data, performing weighted fusion processing on the macro feature data, meso feature data and micro feature data to obtain pipeline feature fusion data.

[0071] Optimized, step S12 is specifically:

[0072] Step S121, obtaining the initial diameter data and centerline position data of the pipeline in the multi-level feature data; using the radial basis function to calculate the actual diameter data of each point along the pipeline; constructing the tangent vector data and the normal vector data according to the centerline position data; calculating the curvature data of each point on the centerline based on the tangent vector data and the normal vector data; combining the actual diameter data and the curvature data to obtain the macro feature data;

[0073] Step S122, obtaining the macroscopic feature data obtained in step S121; calculating the local curvature change rate data according to the curvature data; calculating the diameter change gradient data using a spatial gradient algorithm; combining the local curvature change rate data and the diameter change gradient data to obtain the mesoscopic feature data;

[0074] Step S123, obtaining the mesoscopic feature data obtained in step S122; extracting the surface roughness feature data using a multi-frequency waviness analysis method; constructing the interface feature data according to the material distribution data; combining the surface roughness feature data and the interface feature data to obtain the microscopic feature data.

[0075] Optimized, step 121 is specifically:

[0076] Step S1211, obtaining initial diameter data of the pipeline; selecting a local weighted spline function as a radial basis function; calculating a local weight coefficient of each measuring point; fitting the diameter data using the local weighted spline function; extracting the actual diameter data after fitting;

[0077] Step S1212, obtaining centerline position data; constructing a centerline parameter equation using a cubic spline interpolation method; calculating tangent vector data of each point; calculating normal vector data of each point based on the tangent vector data; obtaining geometric vector data of the pipeline centerline based on the tangent vector data and the normal vector data;

[0078] Step S1213, obtaining the geometric vector data obtained in step S1212; calculating the angle data of adjacent tangent vectors; calculating the curvature data of each point according to the angle data and the arc length data of the pipeline; smoothing the curvature data to obtain optimized curvature data.

[0079] Optimized, step 122 is specifically:

[0080] Step S1221, obtaining the curvature data obtained in step S121; calculating the arc length data between adjacent measuring points; calculating the curvature derivative of each measuring point using the central difference method; calculating the local curvature change rate data based on the curvature derivative and the arc length data;

[0081] Step S1222, obtain the actual diameter data and centerline position data obtained in step S121; construct a three-dimensional grid coordinate system; calculate the differential values ​​of the diameter data in the three coordinate directions; calculate the spatial gradient vector based on the differential values ​​to obtain the diameter change gradient data.

[0082] Optimized, S123 specifically:

[0083] Step S1231, acquiring pipeline surface detection data; performing wavelet decomposition on the detection data to obtain waviness analysis data of different frequency bands; calculating the root mean square roughness value of each frequency band; performing weighted combination on the roughness values ​​of each frequency band to obtain surface roughness characteristic data;

[0084] Step S1232, obtaining pipeline material distribution data; using an adaptive threshold segmentation algorithm to identify the material interface position; calculating the material gradient data at the interface; constructing an interface feature description matrix to obtain interface feature data.

[0085] According to another aspect of the present application, step S2 is further:

[0086] Step S21, obtaining temperature data, vibration data and medium data of pipeline detection, and constructing environmental factor data: according to the environmental factor data, establishing a nonlinear mapping relationship of the impact of environmental factors on detection accuracy; according to the nonlinear mapping relationship, calculating environmental impact weight data;

[0087] Step S22, obtaining the pipeline feature fusion data obtained in step S1 and the environmental impact weight data obtained in step S21; using an environment-aware adaptive partitioning algorithm to process the pipeline feature fusion data, wherein partitioning parameters are adjusted in real time according to the environmental impact weight data; partitioning the pipeline according to the adjusted partitioning parameters to obtain optimized partitioning data;

[0088] Step S23, obtaining the optimized partition data obtained in step S22 and the environmental factor data obtained in step S21; constructing an environmental compensation function to correct the optimized partition data; calculating the time-varying influence of environmental factors according to the environmental compensation function to obtain partition correction data.

[0089] Optimized, step S22 is specifically:

[0090] Step S221, obtaining pipeline feature fusion data; using density clustering method to determine the number of initial partitions; calculating the feature similarity of each area; constructing an initial partition scheme based on the feature similarity;

[0091] Step S222, obtaining the initial partition scheme and environmental impact weight data obtained in step S221; calculating the impact intensity of environmental factors on the boundaries of each partition; constructing a boundary adjustment function; and dynamically optimizing the partition boundaries according to the boundary adjustment function;

[0092] Step S223, obtaining the partition boundary data optimized in step S222; calculating the characteristic uniformity index of each partition; adjusting the partition parameters using an iterative optimization algorithm; generating optimized partition data based on the adjusted partition parameters.

[0093] According to another aspect of the present application, step S3 is further:

[0094] Step S31, determining the data acquisition position according to the partition correction data obtained in step S2; acquiring the pipeline surface image data stream and the ultrasonic data stream; combining the surface image data stream and the ultrasonic data stream to construct a multimodal data acquisition matrix; and obtaining multi-source acquisition data according to the multimodal data acquisition matrix;

[0095] Step S32, acquiring the multi-source acquisition data obtained in step S31; calculating the time delay data between the surface image data stream and the ultrasonic data stream; constructing a space transformation matrix to perform spatial mapping on the multi-source acquisition data; and using a multi-modal data synchronous fusion algorithm to perform spatiotemporal alignment processing on the multi-source acquisition data according to the time delay data and the space transformation matrix to obtain aligned data;

[0096] Step S33, obtaining the aligned data obtained in step S32; constructing adaptive filter parameters to filter the aligned data; using an edge-preserving smoothing algorithm to smooth the filtered data; and obtaining preprocessed data based on the smoothed data.

[0097] Optimized, step S32 is specifically:

[0098] Step S321, obtaining a surface image data stream and an ultrasonic data stream; using a cross-correlation analysis method to calculate time delay characteristics of the two data streams; constructing a time window function; segmenting the data stream according to the time window function to obtain time segmented data;

[0099] Step S322, obtaining the time segment data obtained in step S321; establishing a spatial coordinate transformation matrix; calculating the spatial correspondence between the image feature points and the ultrasonic detection points; and performing spatial registration using a projection transformation method;

[0100] Step S323, obtaining the spatial registration data obtained in step S322; constructing a data fusion weight matrix; using wavelet transform to perform multi-scale decomposition; performing weighted fusion on data of different scales to obtain aligned data.

[0101] Optimized, step S321 is specifically:

[0102] Step S3211, obtaining the sampling timing of the surface image data stream and the ultrasonic data stream; calculating the cross-correlation function value of the two data streams; extracting the cross-correlation peak position; determining the time delay feature according to the peak position;

[0103] Step S3212, obtaining the time delay characteristics obtained in step S3211; constructing a Hanning window function; calculating the optimal window length; performing window segmentation on the data stream to obtain time segmented data.

[0104] Optimized, step S322 is specifically:

[0105] Step S3221, obtaining the coordinates of feature points in the time segment data; constructing an affine transformation matrix; calculating the coordinate mapping relationship; generating space transformation parameters;

[0106] Step S3222, obtain the spatial transformation parameters obtained in step S3221; use the bilinear interpolation method to perform coordinate transformation; calculate the position error after transformation; optimize the transformation parameters to obtain spatial registration data.

[0107] According to another aspect of the present application, step S4 is further:

[0108] Step S41, obtaining the pre-processed data obtained in step S3; constructing a Haar wavelet transform matrix; using the Haar wavelet transform matrix to transform the pre-processed data; and calculating characteristic spectrum data according to the transformation result;

[0109] Step S42, obtaining the feature spectrum data obtained in step S41; inputting the feature spectrum data into the input layer of the hierarchical feature extraction network; extracting features from the input data using a set convolution kernel in the feature extraction layer; processing the extracted features using an activation function; and classifying using a softmax classifier in the classification layer to obtain defect feature data;

[0110] Step S43, obtaining the defect feature data obtained in step S42; constructing a feature fusion matrix; using an attention mechanism to assign weights to the defect feature data; fusing the defect feature data according to the assigned weights to obtain fused feature data.

[0111] Optimized, step S42 is specifically:

[0112] Step S421, obtaining feature map data; constructing an input data standardization layer; calculating data statistical features; normalizing the input data to obtain standardized feature data;

[0113] Step S422, obtaining the standardized feature data obtained in step S421; constructing a multi-scale convolution kernel group; calculating feature mapping; and using an attention mechanism to perform feature enhancement;

[0114] Step S423, obtaining the enhanced feature data obtained in step S422; constructing a feature classifier; calculating the category probability distribution; generating defect feature data.

[0115] Optimized, step S421 is specifically as follows:

[0116] Step S4211, obtain feature map data; calculate the mean and standard deviation of the data; construct batch normalization parameters; perform standardization conversion on the data;

[0117] Step S4212, obtain the standardized data obtained in step S4211; calculate the skewness and kurtosis of the data; construct a data distribution feature vector; generate standardized feature data.

[0118] Optimized, step S422 is specifically:

[0119] Step S4221, obtaining standardized feature data; designing convolution kernel parameters of different scales; calculating convolution feature maps; extracting multi-scale feature mapping data;

[0120] Step S4222, obtain the feature mapping data obtained in step S4221; calculate the channel attention weight; construct the spatial attention matrix; perform weighted enhancement on the features to obtain enhanced feature data.

[0121] Optimized, step S423 is specifically:

[0122] Step S4231, obtaining enhanced feature data; constructing fully connected layer parameters; calculating category scores; generating probability distribution data;

[0123] Step S4232, obtain the probability distribution data obtained in step S4231; design post-processing rules; optimize the classification results; output defect feature data.

[0124] According to another aspect of the present application, step S5 is further:

[0125] Step S51, obtaining the fused feature data obtained in step S4; constructing evaluation indicators of a multi-dimensional credibility evaluation model; performing multi-dimensional credibility calculation on the fused feature data; and obtaining credibility evaluation data according to the calculation results;

[0126] Step S52, obtaining the credibility evaluation data obtained in step S51 and the fused feature data obtained in step S4; using a comprehensive evaluation function to analyze and process the credibility evaluation data and the fused feature data; and obtaining analysis result data according to the analysis and processing results;

[0127] Step S53, obtaining the analysis result data obtained in step S52; extracting the defect location information, type information and size information; integrating the defect location information, type information and size information and the credibility assessment data; and generating a test report.

[0128] Optimized, step S51 is specifically:

[0129] Step S511, obtaining fusion feature data; constructing feature importance scoring criteria; calculating the importance score of each dimension feature; generating feature weight data;

[0130] Step S512, obtaining the feature weight data and fusion feature data obtained in step S511; constructing an evidence theory framework; calculating basic probability distribution; generating credibility assessment data.

[0131] Optimized, step S511 is specifically as follows:

[0132] Step S5111, obtaining fusion feature data; extracting stability index and discrimination index of features; calculating index weights using entropy weight method; generating importance score data;

[0133] Step S5112, obtain the importance score data obtained in step S5111; construct a fuzzy evaluation matrix; calculate the feature membership; generate feature weight data.

[0134] Optimized, step S512 is specifically:

[0135] Step S5121, obtain feature weight data and fusion feature data; establish an evidence framework identification domain; calculate the support of each evidence source; generate basic probability distribution data;

[0136] Step S5122, obtain the basic probability distribution data obtained in step S5121; use the Dempster combination rule; calculate the evidence fusion result; generate credibility assessment data.

[0137] Optimized, step S52 is specifically:

[0138] Step S521, obtaining credibility evaluation data and fusion feature data; constructing a multi-objective decision model; calculating the satisfaction of each evaluation indicator; generating indicator scoring data;

[0139] Step S522, obtain the indicator scoring data obtained in step S521; establish a comprehensive evaluation function; calculate the overall evaluation score; generate analysis result data.

[0140] Optimized, step S521 is specifically as follows:

[0141] Step S5211, obtaining credibility evaluation data and fusion feature data; determining an evaluation index system; calculating the correlation between indicators; generating indicator weight data;

[0142] Step S5212, obtain the indicator weight data obtained in step S5211; construct a fuzzy comprehensive evaluation matrix; calculate the evaluation vector; generate indicator scoring data.

[0143] Optimized, step S522 is specifically as follows: step S5221, obtaining indicator scoring data; constructing evaluation function parameters; calculating weighted comprehensive scores; generating evaluation score data;

[0144] Step S5222, obtain the evaluation score data obtained in step S5221; set the evaluation grade classification standard; determine the evaluation grade; generate analysis result data.

[0145] Optimized, step S53 is specifically:

[0146] Step S531, obtaining analysis result data; building a report template library; extracting key information; generating basic report data;

[0147] Step S532, obtain the basic report data obtained in step S531; construct intelligent description rules; generate report description content; output the test report.

[0148] Optimized, step S531 is specifically as follows:

[0149] Step S5311, obtain analysis result data; establish information extraction rules; filter key data items; generate information summary data;

[0150] Step S5312, obtain the information summary data obtained in step S5311; match the report template; organize the report structure; generate basic report data.

[0151] In another embodiment of the present application, some algorithms are also provided as follows:

[0152] 1. The mathematical model of the multi-level geometric feature adaptive extraction algorithm (MGFAE), including:

[0153] Radial basis function fitting model: D(x)=D0+∑(α i Φ i (x))·W(x);

[0154] Where D0 is the reference diameter, α i is the weight coefficient, Φ i (x) is the radial basis function, W(x) is the local weighting function, and x is the axial position coordinate of the pipeline.

[0155] Local weighting function: W(x)=exp(-||xx i ||² / 2 σ² );

[0156] Among them, x i is the local center point position, and σ is the weighted range parameter used to control the local influence range.

[0157] Curvature calculation model: κ(s)=|r ' (s)×r '' (s)| / |r ' (s)| ³ ;

[0158] Among them, r(s) is the center line parameter equation, s is the arc length parameter, r ' (s) and r '' (s) are the first and second order derivatives, respectively.

[0159] Constraints:

[0160] 0 < α i < 1, to ensure the validity of the weight coefficient; σ> 0, to ensure the rationality of the weighting range; |r ' (s)| ≠0, ensuring the validity of curvature calculation.

[0161] 2. Mathematical model of the environment-aware adaptive partitioning algorithm (EAAP), including:

[0162] Environmental impact weight mapping λ(E) = β·tanh(WE·E + bE);

[0163] Among them, E is the environmental factor matrix, WE is the mapping weight matrix, bE is the bias parameter, and β is the scaling factor.

[0164] Partition boundary adjustment function: B(x)=B0(x) + ΔB(x)·λ(E);

[0165] Among them, B0(x) is the initial boundary position, and ΔB(x) is the boundary adjustment amount.

[0166] Boundary optimization objective function: J=∑w i ||B i - B i -1|| ² +γ·R(B);

[0167] Among them, w i is the boundary segment weight, R(B) is the smoothing regularization term, and γ is the regularization coefficient.

[0168] Iterative update rule: B t +1=B t-η·∇J(B t );

[0169] Among them, η is the learning rate and t is the number of iterations.

[0170] Constraints:

[0171] 0 <β≤ 1, controls the intensity of environmental influence; ||WE|| ≤ C, C is a preset constant, limiting the norm of the weight matrix; 0<η< 1, ensures iterative convergence; γ>0, ensures the regularization effect.

[0172] 3. Mathematical model of multimodal data synchronous fusion algorithm (MDSF), including:

[0173] Time delay feature extraction model: C(τ)=∫I s (t)·I u (t+τ)d t / [√∫I s ² (t)d t ·√∫I u ² (t)d t ];

[0174] Among them, I s (t) is the surface image data stream, I u (t) is the ultrasonic data stream, τ is the time delay parameter, and t is the time variable.

[0175] Space coordinate transformation matrix: P'=H·P + T;

[0176] Among them, P is the original space coordinate, H is the homography matrix, T is the translation vector, and P' is the transformed coordinate.

[0177] Data fusion weight calculation: F(x,t)=∑w i (x,t)·D i (x,t)·exp(-μ||xx i || ² );

[0178] Among them, D i (x,t) is the i-th data source, w i (x, t) is the adaptive weight function, μ is the spatial attenuation coefficient, and x is the spatial position.

[0179] Constraints:

[0180] 0 ≤τ≤τ max , limit the maximum time delay; det(H) ≠ 0, ensure the transformation matrix is ​​reversible; ∑w i(x,t)=1, ensure weight normalization; μ>0, control the spatial influence range.

[0181] 4. Mathematical model of hierarchical feature extraction network (HFEN), including:

[0182] Multi-scale convolution kernel design: K(x,y)=α·exp(-x² / 2σx²-y² / 2σy²)·cos(2πfxx + 2πfyy);

[0183] Among them, σx, σy are scale parameters, fxx, fyy are directional frequencies, and α is the amplitude coefficient.

[0184] Attention mechanism calculation: A(F)=softmax(WQ·F·WK^T)·(WV·F);

[0185] Among them, F is the feature matrix, WQ, WK, WV are the learnable parameter matrices.

[0186] Feature enhancement function: E(F) = F+δ[A(F)⊙G(F)];

[0187] Among them, G(F) is the gating function, δ is the enhancement coefficient, and ⊙ represents the element-by-element product.

[0188] Constraints:

[0189] σx,σy>0, ensuring the effectiveness of the kernel function; ||WQ||,||WK||,||WV||≤M, where M is a preset constant, 0 < δ <1, controlling the enhancement strength.

[0190] 5. Mathematical model of multi-dimensional reliability assessment model (MRAM), including:

[0191] Feature importance score: S(f)=-∑p i ·log(p i )·v i ;

[0192] Among them, p i is the probability distribution of feature f, v i is the performance indicator on the validation set.

[0193] Evidence theory framework: m(A)=[1+exp(-λ·∑w i ·e i (A))]^(-1);

[0194] Among them, m(A) is the basic probability distribution, e i (A) is the source of evidence, w i is the weight of evidence, and λ is the sensitivity parameter.

[0195] Comprehensive evaluation function: R=∑β i ·m i (A)·exp(-ρ·V i );

[0196] Among them, β i To evaluate the dimension weights, V i is the uncertainty quantification value, and ρ is the attenuation coefficient.

[0197] Constraints:

[0198] ∑p i =1, probability normalization; 0≤w i ≤1, range of evidence weight; λ>0, sensitivity of controlling evidence combination; ∑β i =1, the evaluation weight is normalized; ρ>0, the influence of uncertainty is controlled.

[0199] 6. The mathematical model of the edge preserving smoothing algorithm (EPSA) in step S33 includes:

[0200] Adaptive filter function: S(x) = I(x) + k·div[c(||∇I||)·∇I];

[0201] Among them, I(x) is the original data, k is the diffusion coefficient, c(·) is the diffusion function, div is the divergence operator, and ∇ is the gradient operator.

[0202] Diffusion function design: c(||∇I||)=exp(-||∇I||² / K²);

[0203] Among them, K is the gradient threshold parameter, which is used to control the degree of edge preservation.

[0204] Constraints:

[0205] k>0, ensuring the stability of the diffusion process;

[0206] K>0, controls edge detection sensitivity;

[0207] 0≤c(·)≤1, limiting the range of the diffusion function.

[0208] 7. The mathematical model of the comprehensive feature fusion algorithm (CFFA) in step S43 includes:

[0209] Feature fusion matrix: F = W·[αH + (1-α)L]·Q;

[0210] Among them, H is the high-level feature, L is the low-level feature, W is the weight matrix, Q is the quality assessment matrix, and α is the fusion coefficient.

[0211] Attention weight calculation: Q(x)=sigmoid(V·tanh(UH + b));

[0212] Among them, V, U are learnable parameter matrices, and b is the bias term.

[0213] Constraints:

[0214] 0≤α≤1, controls the feature fusion ratio;

[0215] ||W||≤C, C is a preset constant;

[0216] rank(Q) = N, ensuring the effectiveness of quality assessment.

[0217] According to another aspect of the present application, a large-range variable-diameter pipeline detection system is also provided, comprising:

[0218] At least one processor; and a memory in communication with at least one of the processors; wherein the memory stores instructions executable by the processor, and the instructions are used to be executed by the processor to implement the large-range variable diameter pipeline detection method described in any of the above technical solutions.

[0219] According to another aspect of the present application, a large-range variable-diameter pipeline detection device is also provided, comprising:

[0220] Cylinder, driving device, detection device;

[0221] The cylinder is provided with a rotating motor slip ring, which drives the detection device to rotate, and the slip ring prevents the wire from getting tangled during rotation; an end cap is provided at each end of the cylinder, which is used to fix the drive device and the detection device to the cylinder as a whole;

[0222] The driving device is composed of an electric telescopic rod, a sliding guide rod, a bearing seat and a driving wheel device. The driving wheel device is composed of a driving motor, a driving wheel, a motor mounting plate and a support plate. One end of the driving wheel device is connected to a fixed block through a motor mounting plate to form a whole, and the other end is connected to the bearing seat through a support plate. The electric telescopic rod will drive the bearing seat during the telescopic process, and at the same time, the support plate will drive the driving wheel device to expand and contract, thereby adapting to the diameter changes of different pipes.

[0223] The detection device is composed of an electric telescopic rod, a sliding guide rod, a bearing seat and a probe assembly. The probe assembly is composed of a probe frame, a probe, a probe frame mounting plate, a pulley and a support plate. An image capture device and an ultrasonic generator are arranged inside the probe, which are respectively used to obtain image data and ultrasonic waveform data of each area of ​​the pipeline; one end of the detection device is connected to the fixed block through the probe mounting plate to form a whole, and the other end is connected to the bearing seat through the support plate; the electric telescopic rod will drive the bearing seat during the telescopic process, and at the same time, the support plate will drive the detection device to expand and contract, so as to adapt to the diameter changes of different pipelines; the upper part of the probe frame is made of plastic material, and pulleys are installed at both ends, which play a role in protecting the pipe wall during the detection process to avoid scratches and damage.

[0224] It should be noted that the various specific technical features described in the above specific embodiments can be combined in any suitable manner without contradiction. In order to avoid unnecessary repetition, the present invention will not further describe various possible combinations.

Claims

1. A method for detecting a large-range variable-diameter pipeline, characterized in that: The steps include: Step S1, obtaining pipeline CAD design drawing data, using a multi-level geometric feature adaptive extraction algorithm to process the pipeline CAD design drawing data, and extracting multi-level feature data; based on the multi-level feature data, extracting macroscopic feature data, mesoscopic feature data, and microscopic feature data of the pipeline; performing feature fusion on the macroscopic feature data, mesoscopic feature data, and microscopic feature data to obtain pipeline feature fusion data; Among them, the macroscopic feature data includes the diameter change data and centerline curvature data of the pipeline; the mesoscopic feature data includes the local curvature change rate data and the local diameter change gradient data; the microscopic feature data includes the surface roughness feature data and the material interface feature data; Step S2, obtaining the pipeline feature fusion data obtained in step S1; extracting pipeline detection environmental factor data, and constructing environmental impact weight data based on the environmental factor data; performing partition processing using an environment-aware adaptive partitioning algorithm based on the pipeline feature fusion data and the environmental impact weight data to obtain optimized partition data; performing environmental compensation correction based on the optimized partition data and the environmental factor data to obtain partition correction data; Step S3, obtaining the partition correction data obtained in step S2; respectively collecting the pipeline surface image data stream and the ultrasonic data stream to construct a multimodal data acquisition matrix; performing spatiotemporal alignment processing on the multimodal data acquisition matrix to obtain aligned data; performing quality enhancement processing on the aligned data to obtain preprocessed data; Step S4, obtaining the preprocessed data obtained in step S3; performing feature analysis on the preprocessed data using a multi-scale analysis method to obtain feature spectrum data; inputting the feature spectrum data into a hierarchical feature extraction network for processing to obtain defect feature data; performing feature fusion processing on the defect feature data to obtain fused feature data; Step S5, obtaining the fused feature data obtained in step S4; using a multi-dimensional credibility assessment model to perform credibility assessment on the fused feature data to obtain credibility assessment data; performing a comprehensive analysis based on the credibility assessment data to obtain analysis result data; and generating a test report based on the analysis result data.

2. The method for detecting a large-range variable-diameter pipeline according to claim 1, characterized in that: The step S1 is further as follows: Step S11, obtaining pipeline CAD design drawing data, extracting pipeline design parameter data; constructing a feature hierarchy matrix according to the design parameter data, wherein the feature hierarchy matrix includes macro-level features, meso-level features and micro-level features; based on the feature hierarchy matrix, obtaining multi-level feature data; Step S12, obtaining the multi-level feature data in step S11, extracting the macroscopic feature data, mesoscopic feature data and microscopic feature data of the pipeline; wherein extracting the macroscopic feature data includes: extracting the diameter change data and the centerline curvature data of the pipeline; extracting the mesoscopic feature data includes: Calculate local curvature change rate data and local diameter change gradient data; Extracting microscopic feature data, including: extracting surface roughness feature data and material interface feature data; Step S13, obtaining the macro feature data, meso feature data and micro feature data obtained in step S12, and constructing feature importance weight data; according to the feature importance weight data, performing weighted fusion processing on the macro feature data, meso feature data and micro feature data to obtain pipeline feature fusion data.

3. The method for detecting a large-range variable-diameter pipeline according to claim 2, characterized in that: The step S2 is further as follows: Step S21, obtaining temperature data, vibration data and medium data of pipeline detection, and constructing environmental factor data: according to the environmental factor data, establishing a nonlinear mapping relationship of the impact of environmental factors on detection accuracy; according to the nonlinear mapping relationship, calculating environmental impact weight data; Step S22, obtaining the pipeline feature fusion data obtained in step S1 and the environmental impact weight data obtained in step S21; using an environment-aware adaptive partitioning algorithm to process the pipeline feature fusion data, wherein partitioning parameters are adjusted in real time according to the environmental impact weight data; partitioning the pipeline according to the adjusted partitioning parameters to obtain optimized partitioning data; Step S23, obtaining the optimized partition data obtained in step S22 and the environmental factor data obtained in step S21; constructing an environmental compensation function to correct the optimized partition data; calculating the time-varying influence of environmental factors according to the environmental compensation function to obtain partition correction data.

4. The method for detecting a large-range variable-diameter pipeline according to claim 3, characterized in that: The step S3 is further as follows: Step S31, determining the data acquisition position according to the partition correction data obtained in step S2; acquiring the pipeline surface image data stream and the ultrasonic data stream; combining the surface image data stream and the ultrasonic data stream to construct a multimodal data acquisition matrix; and obtaining multi-source acquisition data according to the multimodal data acquisition matrix; Step S32, acquiring the multi-source acquisition data obtained in step S31; calculating the time delay data between the surface image data stream and the ultrasonic data stream; Constructing a spatial transformation matrix to perform spatial mapping on the multi-source collected data; using a multimodal data synchronization fusion algorithm to perform spatiotemporal alignment processing on the multi-source collected data according to the time delay data and the spatial transformation matrix to obtain aligned data; Step S33, obtaining the aligned data obtained in step S32; constructing adaptive filter parameters to filter the aligned data; using an edge-preserving smoothing algorithm to smooth the filtered data; and obtaining preprocessed data based on the smoothed data.

5. The method for detecting a large-range variable-diameter pipeline according to claim 4, characterized in that: The step S4 is further as follows: Step S41, obtaining the pre-processed data obtained in step S3; constructing a Haar wavelet transform matrix; using the Haar wavelet transform matrix to transform the pre-processed data; and calculating characteristic spectrum data according to the transformation result; Step S42, obtaining the feature spectrum data obtained in step S41; inputting the feature spectrum data into the input layer of the hierarchical feature extraction network; extracting features from the input data using a set convolution kernel in the feature extraction layer; processing the extracted features using an activation function; and classifying using a softmax classifier in the classification layer to obtain defect feature data; Step S43, obtaining the defect feature data obtained in step S42; Construct feature fusion matrix; An attention mechanism is used to assign weights to the defect feature data; the defect feature data is fused according to the assigned weights to obtain fused feature data.

6. The method for detecting a large-range variable-diameter pipeline according to claim 5, characterized in that: The step S5 is further as follows: Step S51, obtaining the fused feature data obtained in step S4; constructing evaluation indicators of a multi-dimensional credibility evaluation model; performing multi-dimensional credibility calculation on the fused feature data; and obtaining credibility evaluation data according to the calculation results; Step S52, obtaining the credibility evaluation data obtained in step S51 and the fusion feature data obtained in step S4; The credibility evaluation data and the fusion feature data are analyzed and processed by using a comprehensive evaluation function; and analysis result data are obtained according to the analysis and processing results; Step S53, obtaining the analysis result data obtained in step S52; Extract defect location information, type information and size information; integrate the defect location information, type information and size information and credibility assessment data; generate a test report.

7. Large range variable diameter pipeline detection system, characterized by: include: at least one processor; and a memory communicatively connected to at least one of the processors; wherein, The memory stores instructions executable by the processor, and the instructions are used to be executed by the processor to implement the large-range variable-diameter pipeline detection method according to any one of claims 1 to 6.

8. A large range variable diameter pipeline detection device, characterized in that: include: Cylinder, driving device, detection device; A rotating motor slip ring is installed in the cylinder, and the rotating motor drives the detection device to rotate. The slip ring prevents the wire from getting tangled during rotation; There is an end cap at each end of the cylinder, which is used to fix the driving device and the detection device to the cylinder as a whole; The driving device is composed of an electric telescopic rod, a sliding guide rod, a bearing seat and a driving wheel device. The driving wheel device is composed of a driving motor, a driving wheel, a motor mounting plate and a support plate. One end of the driving wheel device is connected to a fixed block through a motor mounting plate to form a whole, and the other end is connected to the bearing seat through a support plate. The electric telescopic rod will drive the bearing seat during the telescopic process, and at the same time, the support plate will drive the driving wheel device to expand and contract, thereby adapting to the diameter changes of different pipes. The detection device is composed of an electric telescopic rod, a sliding guide rod, a bearing seat and a probe assembly. The probe assembly is composed of a probe frame, a probe, a probe frame mounting plate, a pulley and a support plate. An image capture device and an ultrasonic generator are arranged inside the probe, which are respectively used to obtain image data and ultrasonic waveform data of each area of ​​the pipeline; one end of the detection device is connected to the fixed block through the probe mounting plate to form a whole, and the other end is connected to the bearing seat through the support plate; the electric telescopic rod will drive the bearing seat during the telescopic process, and at the same time, the support plate will drive the detection device to expand and contract, so as to adapt to the diameter changes of different pipelines; the upper part of the probe frame is made of plastic material, and pulleys are installed at both ends, which play a role in protecting the pipe wall during the detection process to avoid scratches and damage.

Images