An insulating intelligent patrol device bonded by adhesive

Abstract

The application belongs to the technical field of circuit detection, and discloses a kind of intelligent inspection device of cementing insulation, the device includes: image intelligent acquisition and multi-parameter collaborative detection module, and the high-definition image of the place of cementing insulation and multidimensional physical parameter are collected;Cementing insulation performance multidimensional evaluation module, evaluation model is constructed based on convolution neural network, realize the intelligent evaluation of cementing insulation performance and generate dynamic analysis curve;Cementing insulation fault diagnosis module, in combination with clustering algorithm and deep learning network, realize cementing insulation fault hidden danger identification and fault type discrimination;Automatic walking drive module is used to adopt track positioning and navigation technology, identifies the position of steel rail, automatically along the predetermined track steady, uniform speed forward, in the process of walking, the collection work of each data is completed simultaneously.The application combines multiple filtering and deep learning denoising reconstruction technology, solves the problem of traditional inspection data single, low precision.

Description

Technical Field

[0001] This invention belongs to, but is not limited to, the field of circuit testing technology, and particularly relates to an intelligent inspection device for adhesive insulation. Background Technology

[0002] In railway signaling systems, track circuits are crucial infrastructure for detecting train occupancy and transmitting operational information. To ensure electrical isolation and maintain mechanical connection between adjacent track circuits, rail insulation is added on-site to delineate track sections. Insulating connection structures are typically installed between rails, with glued-insulated rail joints being a widely adopted form. This structure forms a composite connection through an adhesive layer between the insulating material and the rail, along with a fishplate fastening structure. This ensures that the rails at both ends of the insulated joint maintain necessary mechanical strength while achieving electrical insulation, thus guaranteeing the normal operation of the track circuits. Insulated joints not only serve the connection function of the track structure but also play a vital role in delineating track circuit sections; their reliability directly affects the safe operation of train detection and the signaling system.

[0003] In actual railway lines, glued insulated joints typically use glass resin boards as the insulation layer and are fixed by structural components such as fishplates, insulating gaskets, and insulating sleeves. This structure is prone to wear, cracking, aging, or contamination of the insulation material under long-term train loads, vibrations, and environmental changes. Once the insulation structure is damaged or its insulation performance deteriorates, it can lead to electrical short circuits between track sections, causing misjudgments in the track circuit and affecting train operation safety. Therefore, glued insulated joints have always been considered critical parts of the track circuit and are an area that engineering maintenance and signal maintenance departments need to focus on and regularly inspect.

[0004] Existing railway signaling systems typically include centralized signal monitoring systems to monitor electrical parameters such as voltage and phase at the track circuit receiver in real time, thereby determining the overall operational status of the track circuit. However, these monitoring systems primarily analyze the status of indoor equipment and track circuit electrical parameters, lacking direct, real-time online monitoring methods for the adhesive-bonded insulation structure itself located on outdoor lines. In other words, key parameters such as the insulation performance, structural condition, and contamination status of adhesive-bonded insulation joints are difficult to directly monitor using existing centralized signal monitoring systems.

[0005] Currently, the inspection of adhesive-bonded insulation structures on railway sites mainly relies on manual inspection. Inspectors typically visually check for conductive foreign objects such as iron filings and pieces at the adhesive-bonded insulation joints, the presence of burrs at the rail ends, and wear on the glass resin panels. They then use handheld online insulation resistance testers to perform on-site tests to determine if the insulation resistance meets relevant technical standards. This type of inspection is heavily reliant on manual labor, requiring inspectors to visit each point on-site for testing, resulting in a large workload and low inspection efficiency.

[0006] Furthermore, manual inspection methods are susceptible to various external factors, such as weather conditions, lighting conditions, the stability of the test probes, and the experience level of the inspectors, all of which can lead to errors in the test results. Additionally, due to limitations in manpower and working time, it is difficult to achieve continuous monitoring of the adhesive insulation condition across the entire track, thus failing to grasp the dynamic trends of insulation performance changes over time. If insulation performance gradually deteriorates without timely detection, insulation damage or short circuits will occur, ultimately affecting the normal operation of the track circuit.

[0007] Therefore, current technologies for monitoring the condition of bonded insulation structures still mainly rely on manual inspection and testing. There is a lack of a technical means to automatically inspect and simultaneously complete condition detection and performance evaluation in a track environment, making it difficult to achieve efficient, continuous, and objective monitoring of bonded insulation performance. These issues have become key technical problems that urgently need to be solved in the field of track circuit maintenance. Summary of the Invention

[0008] To address the problems existing in the prior art, the present invention provides an intelligent inspection device for adhesive bonding insulation.

[0009] The present invention is implemented as follows: an intelligent inspection device for adhesive bonding insulation includes an image intelligent acquisition and multi-parameter collaborative detection module, an adhesive bonding insulation performance multi-dimensional evaluation module, an adhesive bonding insulation fault diagnosis module, and an automatic walking drive module.

[0010] The image intelligent acquisition and multi-parameter collaborative detection module, the adhesive insulation performance multi-dimensional evaluation module, and the adhesive insulation fault diagnosis module sequentially establish bidirectional data linkage, forming a closed-loop collaborative mechanism of acquisition-preprocessing-evaluation-diagnosis-feedback optimization. Within the image intelligent acquisition and multi-parameter collaborative detection module, the image acquisition unit, data acquisition unit, and data preprocessing unit achieve synchronous linkage, and the processing results of the data preprocessing unit are fed back to the image acquisition unit and data acquisition unit in real time, dynamically adjusting the acquisition parameters. The adhesive insulation performance multi-dimensional evaluation module constructs an evaluation model based on a convolutional neural network, fusing and analyzing image features with multi-dimensional physical parameters. The adhesive insulation fault diagnosis module combines clustering algorithms and deep learning networks, receiving the feature abstraction results from the adhesive insulation performance multi-dimensional evaluation module to achieve accurate identification of potential faults and rapid judgment of fault types.

[0011] The automatic walking drive module is wirelessly connected to a computer device and uses track positioning and navigation technology to identify the position of the rails and automatically move steadily and at a constant speed along the predetermined track, while simultaneously collecting various data during the walking process.

[0012] Furthermore, the image intelligent acquisition and multi-parameter collaborative detection module includes:

[0013] The image acquisition unit uses a stacked CMOS camera with a global shutter, a large aperture and high resolution lens, and achieves micron-level precision focus adjustment through a stepper motor connection. The camera achieves precise synchronization with the rail being measured through a hardware triggering method.

[0014] The data acquisition unit is a combination of multiple sensors used to collect data in real time, such as the height difference of the rail surface at the adhesive insulation joint, the wear degree of the glass resin plate, the changes in rail gap, insulation resistance, and rail temperature.

[0015] The data preprocessing unit incorporates Gaussian filtering, median filtering, and a deep learning denoising network to remove noise, preserve details, and reconstruct high resolution images acquired by the image acquisition unit, thereby improving image pixel density.

[0016] Furthermore, the global shutter of the image acquisition unit is used to avoid the rolling shutter effect in high-speed motion, the stacked CMOS is used to improve the light sensitivity in low-light environments and reduce image noise, and the large aperture high-resolution lens is used to increase light transmission and ensure image edge sharpness.

[0017] Furthermore, the convolutional neural network of the multi-dimensional evaluation module for adhesive bonding insulation performance is a deep hierarchical feedforward neural network with convolution calculation. By superimposing multiple convolutional layers and nonlinear activation functions, and using learnable convolutional kernels to perform sliding window product summation with the input image, the layer-by-layer abstraction of image features is achieved, thereby completing the analysis of adhesive bonding insulation performance and rail end fat edge status of 25HZ phase-sensitive track circuit.

[0018] Furthermore, the multi-dimensional evaluation module for adhesive bonding insulation performance generates dynamic analysis curves for adhesive bonding insulation performance based on feature abstraction results, realizing automated and intelligent evaluation of adhesive bonding insulation performance, which significantly shortens the time compared to manual inspection.

[0019] Furthermore, the adhesive bonding insulation fault diagnosis module includes:

[0020] The data feature exploration unit uses the K-means clustering algorithm to explore and interpret features of large-scale collected data;

[0021] The feature extraction and analysis unit uses a Long Short-Term Memory (LSTM) network to extract and analyze data features;

[0022] The model training and diagnosis unit trains the model to recognize normal working conditions such as good adhesive insulation appearance, standard rail gaps, no side wear, and no thick edges, and enables the detection of potential faults and the identification of fault types.

[0023] Furthermore, the model training and diagnosis unit incorporates transfer learning technology to enhance the model's generalization ability under different railway line operating conditions, thereby significantly improving the accuracy of intelligent fault diagnosis.

[0024] Furthermore, the device is suitable for the inspection of the track end adhesive bonding insulation of 25HZ phase-sensitive track circuits. The adhesive bonding insulation uses glass resin board as the insulating material and is fixed to the track end by fishplate.

[0025] Another objective of this invention is to provide a method for intelligent inspection of adhesive bonding insulation using the aforementioned intelligent inspection device, the method comprising:

[0026] S1: Using the intelligent image acquisition and multi-parameter collaborative detection module, high-definition images and multi-dimensional physical parameters of the glued insulation are acquired to provide data support for performance evaluation and fault diagnosis.

[0027] S2: Utilizing a multi-dimensional evaluation module for adhesive bonding insulation performance, an evaluation model is constructed based on a convolutional neural network to achieve intelligent evaluation of adhesive bonding insulation performance and generate dynamic analysis curves;

[0028] S3: Using the adhesive bonding insulation fault diagnosis module, combined with clustering algorithms and deep learning networks, we can identify potential adhesive bonding insulation faults and determine the fault type.

[0029] S4: Using the automatic walking drive module with track positioning and navigation technology, the position of the rail is identified, and the vehicle moves steadily and at a constant speed along the predetermined track, while simultaneously collecting various data during the walking process.

[0030] Another object of the present invention is to provide a computer device, the computer device including a memory and a processor, the memory storing a computer program, the computer program being executed by the processor causing the processor to perform the steps of the intelligent inspection method for adhesive insulation; the computer device is placed in a fixed indoor location.

[0031] Based on the above technical solutions and the technical problems solved, the advantages and positive effects of the technical solution to be protected by this invention are as follows:

[0032] First, it achieves high-precision, multi-dimensional synchronous acquisition and intelligent feedback optimization of adhesive bonding insulation inspection data. The image intelligent acquisition and multi-parameter collaborative detection module deeply integrates the image acquisition unit, data acquisition unit, and data preprocessing unit. The data preprocessing unit performs real-time quality assessment of the acquired images and physical parameters. If image blurring, excessive noise, or missing key features are detected, the module dynamically calculates and adjusts acquisition parameters such as exposure time, focal length, and sampling frequency based on preset image quality evaluation indicators and physical parameter stability indicators, ensuring consistently high-quality data acquisition. Simultaneously, this module comprehensively utilizes multiple filtering and deep learning denoising and reconstruction techniques to effectively suppress environmental interference and sensor noise, providing a high-clarity, feature-complete, multi-dimensional data source for subsequent evaluation and diagnostic stages. This mechanism fundamentally solves the problems of single data type, low accuracy, and poor consistency inherent in traditional manual inspection and single-sensor solutions, enabling continuous improvement of raw data quality through closed-loop feedback.

[0033] Secondly, a multi-dimensional evaluation model based on convolutional neural networks was constructed to achieve intelligent quantitative analysis and dynamic monitoring of adhesive bonding insulation performance. The multi-dimensional evaluation module for adhesive bonding insulation performance deeply integrates high-definition image features with multi-dimensional physical parameters such as temperature, humidity, and insulation resistance. It automatically extracts key surface features from the images, such as cracks, corrosion, and rail end thickening, using convolutional neural networks for collaborative analysis with physical parameters. This accurately assesses the current performance status and degradation trend of the adhesive bonding insulation, generating dynamic performance curves and providing intuitive and scientific decision-making basis for maintenance. Compared to the traditional method that relies entirely on manual visual inspection and simple instrument measurements, this evaluation mode significantly improves inspection efficiency, significantly shortens the single inspection cycle, reduces the labor intensity of workers and the safety hazards of working in high-risk line environments, transforming the adhesive bonding insulation status from "periodic check-ups" to "real-time monitoring."

[0034] Third, the fault diagnosis model based on the fusion of cluster analysis and deep learning networks achieves accurate identification of potential faults and rapid discrimination of fault types. The adhesive bonding insulation fault diagnosis module receives the feature abstraction results from the evaluation module, uses the K-means clustering algorithm for unsupervised pre-classification of potential faults, and captures the latent distribution characteristics of the data. Furthermore, it combines a long short-term memory network to model the temporal evolution of performance degradation and introduces transfer learning technology to achieve rapid model adaptation and generalization under limited sample conditions. This diagnostic model achieves an accuracy rate of over 95% in identifying various typical faults and early-stage potential faults in adhesive bonding insulation, effectively eliminating subjective errors, experience dependence, and environmental limitations such as nighttime and severe weather conditions inherent in manual inspections. This significantly improves the real-time performance and reliability of fault warnings, providing strong support for the safe operation of railway lines.

[0035] Fourth, autonomous walking and wireless collaborative operation greatly enhance the level of inspection automation. The autonomous walking drive module uses visual navigation, lidar navigation, or magnetic navigation technology to accurately identify the position of the rails, enabling it to move stably and at a constant speed along a predetermined track autonomously, simultaneously collecting various data during the process without human intervention. It connects wirelessly to computer equipment, forming a complete unmanned intelligent inspection system, significantly increasing the frequency and coverage of inspections, further reducing manpower input, and ensuring the continuity and standardization of operations.

[0036] The overall technical solution of this invention forms a closed-loop collaborative mechanism of acquisition, preprocessing, evaluation, diagnosis, and feedback optimization. The image intelligent acquisition and multi-parameter collaborative detection module, the multi-dimensional evaluation module of adhesive bonding insulation performance, and the adhesive bonding insulation fault diagnosis module establish bidirectional data linkage sequentially. Using a USB 3.0 interface and adhering to the TCP / IP protocol, a timestamp synchronization mechanism achieves efficient data interaction and synchronization. Diagnostic and evaluation results can guide the adjustment of acquisition parameters and model optimization, enabling the system to continuously iterate and evolve during use, maintaining a high degree of adaptability to complex line environments and changing operating conditions. In summary, this invention comprehensively improves data quality, evaluation accuracy, diagnostic accuracy, and automation in the field of intelligent inspection of adhesive bonding insulation, possessing significant industrial application value and promising prospects for promotion. Attached Figure Description

[0037] Figure 1 This is a schematic diagram of the structure of the intelligent inspection device for adhesive bonding insulation provided in an embodiment of the present invention;

[0038] Figure 2 This is a schematic diagram of the image intelligent acquisition and multi-parameter collaborative detection module provided in an embodiment of the present invention;

[0039] Figure 3 This is a schematic diagram of the adhesive insulation fault diagnosis module provided in an embodiment of the present invention;

[0040] Figure 4 This is a flowchart of the intelligent inspection method for adhesive bonding insulation provided in the embodiments of the present invention;

[0041] The diagram shows: 1. Image intelligent acquisition and multi-parameter collaborative detection module; 2. Multi-dimensional evaluation module for adhesive bonding insulation performance; 3. Adhesive bonding insulation fault diagnosis module; 4. Image acquisition unit; 5. Data acquisition unit; 6. Data preprocessing unit; 7. Data feature exploration unit; 8. Feature extraction and analysis unit; 9. Model training and diagnosis unit; 10. Automatic walking drive module. Detailed Implementation

[0042] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0043] like Figure 1 As shown, an embodiment of the present invention provides an intelligent inspection device for adhesive insulation, the device comprising:

[0044] The image intelligent acquisition and multi-parameter collaborative detection module 1 acquires high-definition images and multi-dimensional physical parameters of the glued insulation joint, providing data support for performance evaluation and fault diagnosis.

[0045] The multi-dimensional evaluation module 2 for adhesive bonding insulation performance is connected to the image intelligent acquisition and multi-parameter collaborative detection module 1. Based on the convolutional neural network, an evaluation model is constructed to realize intelligent evaluation of adhesive bonding insulation performance and generate dynamic analysis curves.

[0046] The adhesive bonding insulation fault diagnosis module 3 is connected to the adhesive bonding insulation performance multi-dimensional evaluation module 2. It combines clustering algorithms and deep learning networks to realize the identification of potential adhesive bonding insulation faults and the determination of fault types.

[0047] The automatic walking drive module 10 is wirelessly connected to a computer device. It is used to identify the position of the rail using track positioning and navigation technology, and automatically move along the predetermined track stably and at a constant speed, while simultaneously collecting various data during the walking process.

[0048] The intelligent inspection device for adhesive bonding insulation provided in this embodiment of the invention is installed and runs on the rail line in practical applications. Its overall structure consists of an automatic walking drive module 10, an image intelligent acquisition and multi-parameter collaborative detection module 1, a multi-dimensional evaluation module for adhesive bonding insulation performance 2, and an adhesive bonding insulation fault diagnosis module 3, forming a closed-loop inspection system integrating automatic inspection, data acquisition, performance evaluation, and fault diagnosis. Two-way data linkage is achieved through a USB 3.0 interface, following the TCP / IP protocol and using a timestamp synchronization mechanism to realize data interaction and synchronization.

[0049] The dynamically adjusted acquisition parameters include exposure time, focal length, and sampling frequency. The adjustment is triggered when the data preprocessing unit detects that the acquired image or data is blurry, has excessive noise, or lacks key features. The adjustment algorithm is based on preset image quality evaluation indicators (sharpness indicator, signal-to-noise ratio indicator) and physical parameter stability indicators. The optimal parameter values ​​are calculated through an optimization algorithm (gradient descent algorithm), and then the acquisition parameters are dynamically adjusted.

[0050] The automatic walking drive module 10 is installed at the bottom of the device and includes a track guide wheel assembly, a drive motor assembly, a track positioning sensor, and an attitude stabilization mechanism. During operation, the track guide wheel assembly forms a limiting and guiding structure with the rail surface. The track positioning sensor identifies the rail contour features and running path in real time, and combined with a built-in navigation algorithm, achieves automatic positioning and path maintenance along the rail direction, enabling the device to move stably and at a constant speed along the predetermined route. The drive motor assembly maintains a constant inspection speed through a closed-loop speed control strategy to ensure that each detection module obtains stable and consistent detection conditions during movement, thereby avoiding data errors caused by speed variations in traditional manual inspection. During automatic walking, the device can sequentially pass through each bonded insulation node without manual intervention, achieving continuous inspection.

[0051] When the device moves to the bonded insulation area, the intelligent image acquisition and multi-parameter collaborative detection module 1 is activated simultaneously. This module includes a high-definition industrial camera, a light source compensation unit, an insulation resistance detection unit, and an environmental parameter acquisition unit. The high-definition industrial camera, in conjunction with the light source compensation unit, acquires high-definition images of the surface and edge areas of the bonded insulation structure to identify surface defects such as cracks, peeling, contamination, and structural anomalies. Simultaneously, the insulation resistance detection unit forms a detection loop through conductive contact electrodes on both sides of the rails. When the device moves to the bonded insulation position, it automatically contacts the conductive surface of the rails, applies a stable test voltage to both sides of the rails through a built-in constant test voltage source, and measures the weak leakage current through the bonded insulation structure. Based on the voltage-current relationship, the real-time insulation resistance value is calculated. This detection process is completed while the device is moving and is triggered synchronously with the image acquisition time, ensuring a one-to-one correspondence between electrical parameters and structural images. Furthermore, the environmental parameter acquisition unit simultaneously collects environmental factors affecting insulation performance, such as temperature, humidity, and surface contamination levels, thus forming a multi-dimensional detection data set.

[0052] Subsequently, the acquired image data and electrical parameter data are transmitted in real time to the multi-dimensional evaluation module 2 for adhesive bonding insulation performance. This module constructs an evaluation model based on a convolutional neural network to jointly analyze image feature information, resistance parameters, and environmental parameters, thereby achieving a comprehensive evaluation of the adhesive bonding insulation status. The evaluation results are presented as a time series dynamic analysis curve to reflect the changing trend of adhesive bonding insulation performance.

[0053] The adhesive bonding insulation fault diagnosis module 3 performs in-depth analysis on the data output by the evaluation module. This module uses a clustering algorithm to classify the features of historical detection data and combines a deep learning network to perform pattern matching on the current detection results, thereby identifying potential fault hazards and determining the fault type, such as insulation aging, structural cracking, or a decrease in insulation resistance value caused by contamination.

[0054] Through the coordinated operation of the above structure and steps, this embodiment of the invention realizes continuous inspection, synchronous detection and intelligent analysis of the adhesive-bonded insulation parts of the rails in automatic walking mode, so that mechanical motion, parameter detection and intelligent diagnosis form a unified and coordinated mechanism, which not only improves the inspection efficiency, but also significantly enhances the consistency and reliability of the detection data, thereby effectively improving the monitoring capability of the working status of the adhesive-bonded insulation joint.

[0055] like Figure 2 As shown, the image intelligent acquisition and multi-parameter collaborative detection module includes:

[0056] Image acquisition unit 4 uses a stacked CMOS camera with a global shutter, a large aperture and high resolution lens, and achieves micron-level precision focus adjustment through a stepper motor connection. The camera achieves precise synchronization with the measured rail through hardware triggering. When the global shutter is working, the pixel layer quickly converts light to electricity, and the signal processing circuit layer synchronously and efficiently processes the electrical signal. The two work together to ensure high-speed and high-quality image acquisition.

[0057] Data acquisition unit 5 is a multi-sensor combination used to collect real-time data on the rail surface height difference at the adhesive insulation joint, the wear degree of the glass resin plate, the rail gap change, insulation resistance, and rail temperature. The multi-sensor combination consists of a laser displacement sensor, an infrared temperature sensor, and an electromagnetic induction sensor. The wear degree of the glass resin plate is identified through image recognition.

[0058] The data preprocessing unit 6 incorporates Gaussian filtering, median filtering, and a deep learning denoising network to remove noise, preserve details, and reconstruct high-resolution images acquired by the image acquisition unit, thereby improving image pixel density. The deep learning denoising network is a convolutional autoencoder, U-Net network, or GAN network; the high-resolution reconstruction uses SRCNN or ESRGAN.

[0059] The global shutter of the image acquisition unit 4 is used to avoid the rolling shutter effect in high-speed motion, the stacked CMOS is used to improve the light sensitivity in low light environment and reduce image noise, and the large aperture high resolution lens is used to increase the amount of light and ensure the sharpness of image edges.

[0060] The convolutional neural network of the multi-dimensional evaluation module 2 for adhesive bonding insulation performance is a deep hierarchical feedforward neural network with convolution calculation. It uses multiple convolutional layers superimposed with nonlinear activation functions, and employs a learnable convolutional kernel to perform a sliding window product summation with the input image, achieving layer-by-layer abstraction of image features to complete the analysis of the adhesive bonding insulation performance and rail end fat edge status of the 25Hz phase-sensitive track circuit. The convolutional neural network adopts a 5-layer structure, including 3 convolutional layers and 2 fully connected layers; each convolutional layer uses a 3×3 learnable convolutional kernel, which performs a sliding window product summation with the input image, and is combined with a ReLU nonlinear activation function to achieve layer-by-layer abstraction of image features.

[0061] The multi-dimensional evaluation module 2 for adhesive bonding insulation performance generates dynamic analysis curves for adhesive bonding insulation performance based on feature abstraction results, realizing automated and intelligent evaluation of adhesive bonding insulation performance, which significantly shortens the time compared to manual testing.

[0062] The core innovation of the intelligent inspection device for adhesive bonding insulation in this invention lies in the deep collaborative linkage of various modules and internal units, rather than a simple stacking of independent components or existing technologies. Through the orderly linkage of the image intelligent acquisition and multi-parameter collaborative detection module 1, the multi-dimensional evaluation module for adhesive bonding insulation performance 2, and the adhesive bonding insulation fault diagnosis module 3, combined with the exclusive design and collaborative mechanism of each unit, the automation, intelligence, and high precision of adhesive bonding insulation inspection are achieved. Its overall working principle is as follows, highlighting the integrity and creativity of the technical solution and avoiding the rebuttal of existing technology combinations.

[0063] When the device is in operation, the image intelligent acquisition and multi-parameter collaborative detection module 1 first starts working to complete the basic data acquisition and preprocessing of the glued insulation area, providing standardized and high-quality data support for subsequent evaluation and diagnosis. The image acquisition unit 4, data acquisition unit 5, and data preprocessing unit 6 within this module form a closed-loop collaborative system, rather than working independently. The image acquisition unit 4 uses a stacked CMOS camera with a global shutter, combined with a large aperture and high-resolution lens. It achieves micron-level precision focal length adjustment through a stepper motor connection. During operation, the camera achieves precise synchronization with the measured rail through hardware triggering, effectively avoiding the rolling shutter effect in high-speed inspection. The stacked CMOS improves the light sensitivity in low-light environments and reduces image noise, while the large aperture and high-resolution lens ensures light transmission and image edge sharpness. The three work together to ensure that the acquired high-definition images of the glued insulation area are clear, complete, and distortion-free, laying the foundation for subsequent feature extraction.

[0064] Working synchronously with image acquisition unit 4 is data acquisition unit 5, a multi-sensor combination that collects real-time physical parameters such as rail surface height difference at the adhesive insulation joint, wear degree of glass resin plate, rail gap change, insulation resistance, and rail temperature. This achieves synchronous acquisition of image data and physical parameters, solving the problem of asynchronous image acquisition and parameter detection and data disconnect in existing technologies. This ensures that subsequent evaluations can be based on a comprehensive judgment combining image features and physical parameters, improving evaluation accuracy. After acquisition, data preprocessing unit 6 immediately starts. Its built-in Gaussian filtering, median filtering, and deep learning denoising network work together to remove noise, preserve details, and reconstruct high resolution in the image acquired by image acquisition unit 4. At the same time, it performs outlier removal and standardization processing on the physical parameters acquired by data acquisition unit 5, improving image pixel density and the reliability of parameter data. The preprocessed image data and physical parameters are synchronously transmitted to the multi-dimensional evaluation module 2 of adhesive insulation performance through a dedicated data interface.

[0065] The multi-dimensional evaluation module 2 for adhesive bonding insulation performance establishes a stable data connection with the image intelligent acquisition and multi-parameter collaborative detection module 1. After receiving preprocessed image data and physical parameters, it performs its work through an evaluation model built on a convolutional neural network. This convolutional neural network is a deep hierarchical feedforward neural network with convolutional calculations. By superimposing multiple convolutional layers and nonlinear activation functions, it uses a learnable convolutional kernel and the input image to perform sliding window product summation, achieving layer-by-layer abstraction of image features. At the same time, the multi-dimensional physical parameters acquired by the data acquisition unit 5 are integrated into the evaluation model as auxiliary features, breaking the limitation of existing technologies that rely solely on a single image or single parameter for evaluation, and realizing a comprehensive analysis of the adhesive bonding insulation performance and rail end fat edge status of the 25Hz phase-sensitive track circuit.

[0066] The multi-dimensional evaluation module 2 for adhesive bonding insulation performance generates dynamic analysis curves for adhesive bonding insulation performance based on feature abstraction results and multi-parameter fusion analysis. This enables automated and intelligent evaluation of adhesive bonding insulation performance, significantly shortening the time required for manual inspection and improving evaluation accuracy by more than 30% compared to existing single-detection methods. After evaluation, module 2 synchronously transmits the evaluation results and related feature data to the adhesive bonding insulation fault diagnosis module 3. This module works in conjunction with the multi-dimensional evaluation module 2, combining clustering algorithms and deep learning networks to further analyze abnormal features in the evaluation results, thereby identifying potential adhesive bonding insulation faults and determining fault types.

[0067] Throughout the entire process, the three core modules form a complete closed loop of acquisition, preprocessing, evaluation, and diagnosis. The collaboration between these modules and their internal units is not a simple sequential process, but rather a mutual cooperation and feedback mechanism: the synchronous collaboration between image acquisition unit 4 and data acquisition unit 5 ensures data integrity; the processing results of data preprocessing unit 6 directly affect the accuracy of evaluation module 2; the feature data of evaluation module 2 provides accurate basis for diagnosis module 3; and the fault identification results of diagnosis module 3 can be fed back to acquisition module 1 to optimize subsequent acquisition parameters, forming a dynamically optimized inspection mechanism. The specific feedback includes: fault type identification to clarify the type of fault (e.g., delamination, aging); fault severity level (divided into minor, moderate, and severe levels); and acquisition parameter adjustment logic. Based on the fault type and severity, combined with a preset expert rule base and historical data statistical analysis results, the logic determines the adjustment direction and magnitude of parameters such as resolution, exposure time, and focus mode of the image acquisition unit, and sampling frequency and range of the data acquisition unit. This optimizes subsequent acquisition parameters, ensuring that the acquired data more accurately reflects the true state of the adhesive bonding insulation.

[0068] In the above working mechanism, the collaborative design of each module, the synchronous acquisition and fusion analysis of images and multiple parameters, and the combined application of convolutional neural networks and clustering algorithms are all organic wholes that cannot be achieved through simple combinations of existing technologies. Its overall synergy solves the technical pain points of low efficiency, poor accuracy, and low automation in existing adhesive bonding insulation inspection, highlighting the creativity of the technical solution of the present invention embodiment and effectively avoiding the possibility of rebuttal by existing technology combinations.

[0069] like Figure 3 As shown, the adhesive bonding insulation fault diagnosis module 3 includes:

[0070] Data feature exploration unit 7 uses the K-means clustering algorithm to explore and interpret features of large-scale collected data;

[0071] Feature extraction and analysis unit 8 uses a Long Short-Term Memory (LSTM) network to extract and analyze data features;

[0072] Model training and diagnosis unit 9 trains the model to recognize normal working conditions such as good appearance of adhesive insulation, standard rail gaps, no side wear, and no burrs, and enables the detection of potential faults and the identification of fault types.

[0073] The model training and diagnosis unit 9 introduces transfer learning technology to improve the model's generalization ability under different railway line conditions, thereby significantly improving the accuracy of intelligent fault diagnosis.

[0074] The device is suitable for inspecting the adhesive bonding insulation of 25Hz phase-sensitive track circuits. Glass resin board is used as the insulating material at the adhesive bonding insulation point, and it is fixed to the rail end by fishplate.

[0075] like Figure 4 As shown in the embodiment of the present invention, the method for intelligent inspection of adhesive insulation using the intelligent inspection device for adhesive insulation includes:

[0076] S1: Using the intelligent image acquisition and multi-parameter collaborative detection module, high-definition images and multi-dimensional physical parameters of the glued insulation are acquired to provide data support for performance evaluation and fault diagnosis.

[0077] S2: Utilizing a multi-dimensional evaluation module for adhesive bonding insulation performance, an evaluation model is constructed based on a convolutional neural network to achieve intelligent evaluation of adhesive bonding insulation performance and generate dynamic analysis curves;

[0078] S3: Using the adhesive bonding insulation fault diagnosis module, combined with clustering algorithms and deep learning networks, we can identify potential adhesive bonding insulation faults and determine the fault type.

[0079] S4: Using the automatic walking drive module with track positioning and navigation technology, the position of the rail is identified, and the vehicle moves steadily and at a constant speed along the predetermined track, while simultaneously collecting various data during the walking process.

[0080] Specific methods for insulation resistance testing:

[0081] A specially designed electromagnetic induction sensor is installed on the intelligent inspection device for adhesive bonding insulation. This sensor includes a transmitting coil and a receiving coil. The transmitting coil is energized with an alternating current of a specific frequency to generate an alternating magnetic field; the receiving coil is used to detect changes in the magnetic field caused by changes in the conductivity of the adhesive bonding insulation part, and transmits the signal to the data processing unit for analysis and calculation to obtain the insulation resistance value.

[0082] Specific parameters:

[0083] Alternating current frequency range

[0084] Recommended frequency range: 1kHz to 100kHz (mid-frequency range). This range balances electromagnetic field penetration depth and signal detection sensitivity, avoiding insufficient penetration due to too low a frequency or signal attenuation due to too high a frequency.

[0085] Selection criteria: The optimal frequency needs to be determined through experimentation based on the material properties, dimensions, and required depth of the adhesive insulation component. For example, a higher frequency (e.g., 50kHz-100kHz) can be selected for thin-layer insulation materials; while a lower frequency (e.g., 1kHz-10kHz) is required for thick-layer or composite materials.

[0086] Geometric parameters of the transmitting and receiving coils

[0087] Number of turns:

[0088] Transmitting coil: The recommended number of turns is 800-1500, depending on the required magnetic field strength and coil size. For example, if a stronger magnetic field is required, 1200 turns can be selected; if space is limited, the number of turns can be reduced to 800-1000 turns.

[0089] Receiver coil: 500-1000 turns are recommended to balance signal sensitivity and noise suppression. If there is significant interference in the detection environment, the number of turns can be increased to improve the signal-to-noise ratio.

[0090] diameter:

[0091] Transmitting coil: The recommended diameter is 20-50mm, depending on the size of the detection area and the required uniformity of the magnetic field distribution. If the detection area is large, a larger diameter (e.g., 40-50mm) can be selected; if fine local detection is required, a smaller diameter (e.g., 20-30mm) should be selected.

[0092] Receiver coil: The diameter should be slightly smaller than that of the transmitter coil to reduce direct coupling interference. For example, if the transmitter coil diameter is 40mm, the receiver coil diameter can be selected as 30-35mm.

[0093] spacing:

[0094] The recommended spacing between the transmitting and receiving coils is 10-30mm, depending on the magnetic field attenuation characteristics and detection sensitivity requirements. Too small a spacing may lead to direct coupling interference; too large a spacing will weaken the signal strength. The optimal spacing can be determined experimentally.

[0095] Calibration method between insulation resistance value and induced signal

[0096] Calibration method:

[0097] Experimental preparation: Prepare a series of adhesive-bonded insulation samples with known insulation resistance values, ensuring that the sample size, material and other parameters are consistent with the actual test object.

[0098] Data acquisition: Each sample is tested using a patrol inspection device, and the induced electromotive force value in the receiving coil is recorded.

[0099] Calibration curve plotting: Plot the calibration curve with insulation resistance as the x-axis and induced electromotive force as the y-axis. Obtain the mathematical expression between induced electromotive force and insulation resistance value through curve fitting (such as linear regression, polynomial regression, etc.).

[0100] Model validation: The calibration model is validated using another set of samples with known insulation resistance values ​​to ensure the model's accuracy and reliability.

[0101] In embodiments of the present invention, the intelligent inspection method for bonded insulation achieves automatic inspection and intelligent evaluation of railway bonded insulation structures through the coordinated operation of the device structure, detection mechanism, and intelligent analysis algorithm. During operation, the method relies on an intelligent inspection device for bonded insulation, and a continuous and coordinated workflow is formed between the automatic walking drive module, the intelligent image acquisition and multi-parameter collaborative detection module, the multi-dimensional evaluation module for bonded insulation performance, and the bonded insulation fault diagnosis module, thereby constructing a complete track site inspection and analysis mechanism.

[0102] In practical implementation, the automatic walking drive module first identifies the geometric contour and direction of travel of the rails using track positioning and navigation technology. A guide wheel assembly establishes a limiting and guiding relationship with the rails, and the drive motor outputs stable power, enabling the inspection device to move stably and at a constant speed along the predetermined track path. A speed closed-loop control mechanism maintains a constant operating speed during the inspection device's movement, ensuring stable detection conditions for each detection module even while in motion, thereby guaranteeing the consistency and comparability of data collected from different detection points. During its movement, the device sequentially passes through each bonded insulation location, achieving continuous inspection and synchronous detection.

[0103] When the device moves to the bonded insulation area, the intelligent image acquisition and multi-parameter collaborative detection module automatically starts. This module acquires high-definition image information of the surface of the bonded insulation structure and the rail end area through the high-definition image acquisition unit, and simultaneously acquires relevant physical parameter data. At the same time, the electromagnetic induction sensor installed on the device initiates the insulation resistance detection process. The electromagnetic induction sensor includes a transmitting coil and a receiving coil. Under the drive of the control circuit, an alternating current of a specific frequency is passed through the transmitting coil, thereby forming a stable alternating magnetic field near the bonded insulation. When there is a conductive path or a change in insulation performance at the bonded insulation, the magnetic field distribution will change accordingly. This change is sensed by the receiving coil and converted into an electrical signal. The data processing unit analyzes the amplitude and phase characteristics of the received signal and calculates the corresponding insulation resistance parameters through a preset calculation model. Because this detection process uses a non-contact electromagnetic induction method, the detection can be completed during device movement, thus avoiding the measurement instability problems caused by traditional probe contact methods and achieving synchronous detection with image acquisition.

[0104] Subsequently, the acquired image data, insulation resistance parameters, and other test data are uniformly transmitted to the multi-dimensional evaluation module for adhesive bonding insulation performance. This module constructs an evaluation model based on convolutional neural networks to perform joint feature extraction and analysis of image feature information and electrical parameters, thereby achieving a comprehensive evaluation of the adhesive bonding insulation structure status. The evaluation results are generated as a dynamic analysis curve in time series form to reflect the changing trend of adhesive bonding insulation performance in different inspection cycles.

[0105] After obtaining the performance evaluation results, the adhesive bonding insulation fault diagnosis module further performs in-depth analysis of the data. This module uses a clustering algorithm to classify features from historical detection data and combines this with a deep learning network for pattern recognition, thereby identifying potential faults and determining the type of fault in the current detection results. For example, when an abnormal trend in insulation resistance is detected and matches cracks or contamination features in the image, the system can determine it as a potential insulation failure and generate corresponding fault diagnosis results.

[0106] Furthermore, in embodiments of the present invention, the inspection method can also be executed by a computer device. This computer device includes a memory and a processor. The memory stores a computer program for implementing the steps of the above-described inspection method. When the processor executes the computer program, it can complete functions such as data acquisition and control, signal analysis, performance evaluation, and fault diagnosis according to a predetermined process.

[0107] Through the above embodiments, this invention integrates automatic walking inspection, non-contact insulation resistance detection, multi-parameter data acquisition, and intelligent analysis algorithms into a unified collaborative design. This unifies the mechanical inspection process with data detection and intelligent evaluation, thereby achieving continuous inspection and intelligent status monitoring of bonded insulation structures, improving inspection efficiency and detection reliability. It should be noted that the embodiments of this invention can be implemented using hardware, software, or a combination of both. The hardware portion can be implemented using dedicated logic; the software portion can be stored in memory and executed by an appropriate instruction execution system, such as a microprocessor or dedicated hardware. Those skilled in the art will understand that the above-described devices and methods can be implemented using computer-executable instructions and / or included in processor control code, for example, on a carrier medium such as a disk, CD, or DVD-ROM, a programmable memory such as read-only memory (firmware), or a data carrier such as an optical or electronic signal carrier. The device and its modules of the present invention can be implemented by hardware circuits such as very large-scale integrated circuits or gate arrays, semiconductors such as logic chips, transistors, or programmable hardware devices such as field-programmable gate arrays, programmable logic devices, etc., or by software executed by various types of processors, or by a combination of the above hardware circuits and software, such as firmware.

[0108] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications, equivalent substitutions, and improvements made by those skilled in the art within the scope of the technology disclosed in the present invention, and within the spirit and principles of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. An intelligent inspection device for adhesive-bonded insulation, characterized in that, It includes an image intelligent acquisition and multi-parameter collaborative detection module, a multi-dimensional evaluation module for adhesive bonding insulation performance, an adhesive bonding insulation fault diagnosis module, and an automatic walking drive module; The image intelligent acquisition and multi-parameter collaborative detection module, the adhesive bonding insulation performance multi-dimensional evaluation module, and the adhesive bonding insulation fault diagnosis module sequentially establish bidirectional data linkage, forming a closed-loop collaborative mechanism of acquisition-preprocessing-evaluation-diagnosis-feedback optimization; the bidirectional data linkage is achieved by using a USB 3.0 interface, following the TCP / IP protocol, and employing a timestamp synchronization mechanism to realize data interaction and synchronization; The image acquisition unit, data acquisition unit, and data preprocessing unit within the intelligent image acquisition and multi-parameter collaborative detection module are synchronized and linked. The processing results of the data preprocessing unit are fed back to the image acquisition unit and the data acquisition unit in real time, dynamically adjusting the acquisition parameters. The dynamically adjusted acquisition parameters include exposure time, focal length, and sampling frequency. The adjustment is triggered when the data preprocessing unit detects that the acquired image or data is blurry, has excessive noise, or lacks key features. The adjustment algorithm is based on preset image quality assessment indicators and physical parameter stability indicators. The optimal parameter values ​​are calculated through optimization algorithms, and the acquired parameters are dynamically adjusted. The multi-dimensional evaluation module for adhesive bonding insulation performance is based on a convolutional neural network to construct an evaluation model, which integrates image features with multi-dimensional physical parameters for analysis. The adhesive bonding insulation fault diagnosis module combines clustering algorithms and deep learning networks to receive the feature abstraction results from the multi-dimensional evaluation module for adhesive bonding insulation performance, thereby achieving accurate identification of potential faults and rapid judgment of fault types. The automatic walking drive module is wirelessly connected to a computer device and is used to identify the position of the rail using visual navigation, lidar navigation or magnetic navigation technology, and automatically move steadily and at a constant speed along the predetermined track, while simultaneously collecting various data during the walking process.

2. The intelligent inspection device for adhesive bonding insulation according to claim 1, characterized in that, The image acquisition unit uses a stacked CMOS camera with a global shutter, paired with a large aperture and high resolution lens. It achieves micron-level precision focal length adjustment through a stepper motor connection. The camera achieves precise synchronization with the measured rail through hardware triggering. When the global shutter is working, the pixel layer quickly converts light to electricity, and the signal processing circuit layer synchronously and efficiently processes the electrical signal. The two work together to ensure high-speed and high-quality image acquisition.

3. The intelligent inspection device for adhesive bonding insulation according to claim 1, characterized in that, The data acquisition unit is a multi-sensor combination used to collect real-time data on the rail surface height difference at the adhesive insulation joint, the wear degree of the glass resin plate, the rail gap change, the insulation resistance, and the rail temperature. The multi-sensor combination consists of a laser displacement sensor, an infrared temperature sensor, and an electromagnetic induction sensor. The wear degree of the glass resin plate is identified through image recognition.

4. The intelligent inspection device for adhesive bonding insulation according to claim 1, characterized in that, The data preprocessing unit incorporates Gaussian filtering, median filtering, and a deep learning denoising network to remove noise, preserve details, and reconstruct high resolution images acquired by the image acquisition unit. Simultaneously, it performs outlier removal and standardization on the physical parameters acquired by the data acquisition unit. The deep learning denoising network is a convolutional autoencoder, U-Net network, or GAN network. The high resolution reconstruction uses SRCNN or ESRGAN.

5. The intelligent inspection device for adhesive bonding insulation according to claim 1, characterized in that, The convolutional neural network is a deep hierarchical feedforward neural network with convolutional computation. It achieves layer-by-layer abstraction of image features by superimposing multiple convolutional layers and non-linear activation functions, and using learnable convolutional kernels to perform sliding window product summation with the input image. The convolutional neural network adopts a 5-layer structure, including 3 convolutional layers and 2 fully connected layers. Each convolutional layer uses a 3×3 learnable convolutional kernel, which performs sliding window product summation with the input image and is combined with the ReLU non-linear activation function to achieve layer-by-layer abstraction of image features.

6. A method for intelligent inspection of adhesive-bonded insulation, characterized in that, The intelligent inspection device for adhesive bonding insulation according to any one of claims 1 to 5 comprises the following steps: Step 1, the intelligent image acquisition and multi-parameter collaborative detection module is activated, and the image acquisition unit and the data acquisition unit work synchronously to acquire high-definition images and multi-dimensional physical parameters of the adhesive bonding insulation, respectively; Step 2, the data preprocessing unit preprocesses the acquired images and physical parameters; Step 3, the multi-dimensional evaluation module for adhesive bonding insulation performance receives the preprocessed data, performs comprehensive analysis through the evaluation model, and generates a dynamic analysis curve for adhesive bonding insulation performance; Step 4, the adhesive bonding insulation fault diagnosis module receives the evaluation results and performs fault hazard identification and fault classification. Step 5: The adhesive bonding insulation fault diagnosis module feeds back the fault identification results to the image intelligent acquisition and multi-parameter collaborative detection module to optimize subsequent acquisition parameters. The specific feedback includes: fault type identification, used to clarify what kind of fault it is, such as delamination or aging; fault severity level, divided into minor, moderate, and severe levels; acquisition parameter adjustment logic, which, based on the fault type and severity, combined with the preset expert rule base and historical data statistical analysis results, determines the resolution, exposure time, and focus mode of the image acquisition unit, as well as the direction and magnitude of the adjustment of the sampling frequency and range of the data acquisition unit, thereby optimizing subsequent acquisition parameters.

7. The intelligent inspection method for adhesive bonding insulation according to claim 6, characterized in that, In step one, when the image acquisition unit acquires images, it achieves precise synchronization with the rail being tested through hardware triggering to avoid the jelly effect during high-speed inspection.

8. The intelligent inspection method for adhesive bonding insulation according to claim 6, characterized in that, In step two, the data preprocessing unit processes the image to increase the image pixel density and processes the physical parameters to improve the reliability of the parameter data.

9. The intelligent inspection method for adhesive bonding insulation according to claim 6, characterized in that, In step three, the multi-dimensional evaluation module for adhesive bonding insulation performance incorporates the multi-dimensional physical parameters acquired by the data acquisition unit as auxiliary features into the evaluation model, thereby achieving a comprehensive analysis of the adhesive bonding insulation performance and rail end fat edge status of the 25Hz phase-sensitive track circuit.

10. The intelligent inspection method for adhesive-bonded insulation according to claim 6, characterized in that, In step three, the multi-dimensional evaluation module for adhesive bonding insulation performance enables automated and intelligent evaluation of adhesive bonding insulation performance.

Images