Corrugated pipe online forming abnormity judgment method and system based on image recognition

Abstract

The invention discloses a corrugated pipe on-line forming abnormity judgment method and system based on image recognition, and relates to the technical field of image data processing, the system is composed of a plurality of functional modules, and the system comprises a data acquisition preprocessing module for obtaining corrugated pipe image data and equipment vibration data; the feature extraction module is used for segmenting the corrugated pipe image data by utilizing a synchronization mechanism; based on the segmented result, input features are extracted, and the input features comprise head and tail image features and pipeline image features; acquiring multiple groups of input features, and training a distribution model including a lightweight model and a high-precision model; extracting vibration characteristics based on the equipment vibration data; the abnormity judgment module is used for judging head and tail image abnormity and pipeline image abnormity in the input characteristics according to the distribution model, and outputting an abnormal position area; vibration mapping formed by the vibration characteristics and the abnormal position area is obtained, and vibration feedback is established.

Description

Technical Field

[0001] The present invention relates to the technical field of image data processing, and specifically to an online forming abnormality judgment method and system for corrugated pipes based on image recognition. Background Art

[0002] With the development of technology, the continuous hydroforming technology has gradually matured; this technology realizes the continuous processing of tube blanks through multi-station molds to form waves one by one, in cooperation with the advancement and retraction of the mandrel and the sealing system. The introduction of mechanical rolling forming and hydroforming equipment has improved production efficiency and met the requirements for large-diameter and high-wave-depth corrugated pipes.

[0003] Traditional online forming abnormality solutions for corrugated pipes usually use single-perspective static shooting and rely mostly on single-modal sensors, making it difficult to cover surface scratches, corrugation deformation, and periodic abnormalities on the corrugated pipes. The recognition algorithm based on traditional machine vision uses a single-model serial processing architecture and needs to go through multiple processing steps such as image noise reduction (taking about 20 ms), jitter compensation (15 ms), and feature extraction (18 ms), with the single processing time exceeding 50 ms; while modern high-speed production lines can have a throughput of up to 1200 pieces per minute (the single-piece processing time limit ≤ 40 ms), and this mismatch in time windows directly leads to a loss of control in the online detection speed ratio of 3% - 5%. Summary of the Invention

[0004] (I) Technical Problems to be Solved Aiming at the deficiencies of the prior art, the present invention provides an online forming abnormality judgment method and system for corrugated pipes based on image recognition, and solves the problems raised in the background art.

[0005] (II) Technical Solutions To achieve the above objectives, the present invention is realized through the following technical solutions: An online forming abnormality judgment system for corrugated pipes based on image recognition, comprising: A data acquisition and preprocessing module, which acquires corrugated pipe image data and equipment vibration data; A feature extraction module, which uses a synchronization mechanism to segment the corrugated pipe image data; based on the segmented results, extracts input features, and the input features include head and tail image features and pipe image features; obtains multiple groups of input features and trains a distribution model, and the distribution model includes a lightweight model and a high-precision model; based on the equipment vibration data, extracts vibration features; An abnormality judgment module, which judges head and tail image abnormalities and pipe image abnormalities in the input features according to the distribution model, and outputs the abnormal position area; obtains the vibration mapping formed by the vibration features and the abnormal position area, and establishes a vibration feedback; A real-time feedback module, which recognizes the online forming abnormality of the corrugated pipe based on the vibration feedback.

[0006] Further, the process of obtaining bellows image data and equipment vibration data is as follows: Using a synchronization mechanism, perform a three-dimensional circular scan on the surface of the bellows to obtain a multi-spectral image, and perform filtering processing on the spectral image to obtain bellows image data; Using a dynamic signal acquisition card, collect the vibration signals during the operation of the bellows production line, and eliminate the baseline drift of the vibration signals to obtain equipment vibration data.

[0007] Further, segment the bellows image data: According to the speed of the bellows production line, set the segmentation points of the bellows image data, map the segmentation points into the synchronization mechanism, and use the synchronization mechanism to segment the bellows image data.

[0008] Further, extract the head and tail image features: Adopt the Canny algorithm to extract the edge contour of the bellows, respectively obtain the horizontal and vertical three-dimensional matrix information of the bellows image data, and calculate the gradients and gradient amplitudes of the Sobel operators in the horizontal and vertical directions. The formulas are as follows: In the formula, I is the input bellows image data; is the gradient calculated by the Sobel operator in the horizontal direction; is the gradient calculated by the Sobel operator in the vertical direction; According to the gradients calculated by the Sobel operators in the horizontal and vertical directions, calculate the gradient amplitude and gradient direction. The formulas are as follows: In the formula, is the amplitude; is the direction; Separate the bellows area from the background area and perform parameter verification; extract the geometric features of the head and tail ends, and perform texture feature verification and defect preliminary screening by deploying the MobileNetV3 model.

[0009] Further, pipeline image features: Through the bellows image data, identify the ripple period and structural continuity of the bellows. According to the ripple period and structural continuity of the bellows, calculate the mean and standard deviation of the distances between adjacent wave peaks and wave valleys. The formulas are as follows: In the formula, is the mean ripple period; is the standard deviation; is the adjacent peak / trough spacing; N is the total number of detection cycles; i is the current cycle number, i = 1, 2, 3... N; Using the kurtosis index to evaluate the regularity of the corrugation shape, performing a fast Fourier transform, extracting the main frequency component and the proportion of its harmonic energy, and quantifying the corrugation spacing. The formula is: In the formula, is the energy quantization value; is the spectrum amplitude; f is the spectrum amplitude input parameter; M is the total harmonic order; is the main frequency energy ratio; n is the current harmonic order, n = 1, 2, 3... M; Vibration characteristics: Extract vibration characteristics by multi-stage collaborative processing of vibration data representation and dimensionality reduction.

[0010] Furthermore, distribution model training: Lightweight model: Obtain the first and last contours of the bellows by the Canny algorithm, and deploy the lightweight MobileNetV3 model to verify the surface geometric parameters of the bellows; High-precision model: Use wavelet transform to decompose the edge details and low-frequency background components of the bellows image data, optimize through the Retinex algorithm to obtain the contrast of uneven illumination areas, and combine with the frequency-domain high-pass filtering algorithm to strengthen the defect edges and detect robustness.

[0011] Furthermore, abnormal first and last images: Based on the YOLOv5-tiny lightweight detector, locate the first and last image areas, adopt the random cropping strategy of SAFE to crop the bellows image data, retain the geometric features of the end faces, and perform weighted fusion of the geometric parameters and texture features of the first and last areas; Abnormal pipeline images: Use YOLOv5-tiny as a fast positioning model for the first and last areas; divide the cells along the pipeline axis, count the histogram of gradient directions in each cell, and capture local deformation features such as dents and bulges; Introduce the LSTM network to process consecutive frame images, capture the characteristics of defects changing dynamically with the pipeline, and obtain the false detection rate caused by vibration in static images.

[0012] Furthermore, establish vibration feedback: At the detection points corresponding to the abnormal areas, compare the FRF curves of the experiment and the simulation, analyze the amplitude difference and peak frequency shift; generate MAC matrix and FRAC curve quantization indicators to show the matching degree between the vibration characteristics and the abnormal areas; feedback the mapping results to the simulation model.

[0013] An online forming anomaly judgment method for bellows based on image recognition includes the following steps: Step 1: Obtain bellows image data and equipment vibration data; Step 2: Use a synchronization mechanism to segment the bellows image data; based on the segmented results, extract input features, where the input features include head and tail image features and pipe image features; obtain multiple sets of input features and train a distribution model, where the distribution model includes a lightweight model and a high-precision model; based on the equipment vibration data, extract vibration features; Step 3: According to the distribution model, judge the head and tail image anomalies and pipe image anomalies in the input features, and output the abnormal position area; obtain the vibration mapping formed by the vibration features and the abnormal position area, and establish a vibration feedback; Step 4: Based on the vibration feedback, identify the anomalies in the online forming of the bellows.

[0014] (III) Beneficial Effects The present invention provides a method and system for judging anomalies in the online forming of bellows based on image recognition, having the following beneficial effects: (1) This solution adopts a segmented image recognition architecture and an intelligent algorithm optimization strategy, decomposes the bellows forming process into a two-stage processing mechanism, constructs a feature analysis model based on probability distribution, extracts regionalized features of forming parameters in the primary recognition stage, and combines with a deep convolutional network to achieve rapid comparison of morphological parameters, effectively breaking through the computational efficiency bottleneck of traditional single-model processing of high-dimensional data; reducing computational resource consumption, with the real-time monitoring response speed reaching the millisecond level, while ensuring the anomaly detection rate, successfully realizing the dynamic compensation and quality traceability of forming process parameters, and providing a reliable intelligent quality control solution for continuous production lines.

[0015] (2) This solution adopts vibration data acquisition technology combined with an image detection method to identify and analyze abnormal vibrations in the bellows production process from multiple dimensions; by fusing vibration signals and real-time image processing, it can not only accurately capture subtle forming anomalies, but also effectively distinguish the vibration modes and fault modes under normal working conditions, improving the accuracy of judging abnormal situations occurring in the online forming process of bellows. Description of the Drawings

[0016] Figure 1 It is a schematic diagram of the system flow of the present invention; Figure 2 It is a schematic diagram of the overall method of the present invention. Specific Embodiments

[0017] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0018] Embodiment 1: Please refer to Figure 1 , this embodiment provides an online forming abnormality judgment system for corrugated pipes based on image recognition. The system includes: A data acquisition and preprocessing module that acquires corrugated pipe image data and equipment vibration data; Corrugated pipe image data: Use a high-resolution camera to perform a three-dimensional circular scan on the surface of the corrugated pipe; during the three-dimensional circular scan of the surface of the corrugated pipe under the illumination of a multi-spectral light source, perform dynamic light compensation to obtain a spectral image and retain enough details; use a synchronization mechanism to control the shooting timing of the high-resolution camera to avoid motion blur; through preprocessing of the spectral image, obtain the corrugated pipe image data; High-resolution camera: Adopt the Alvium G1 series industrial cameras (gigabit network interface), build a circular imaging array at the end of the corrugated pipe production line, carry 6 cameras (equally spaced at 60°) through a circular stainless steel bracket, and axially arrange an angle-adjustable pan-tilt to achieve 360° dead-angle-free coverage of the forming surface of the corrugated pipe; during the acquisition of corrugated pipe image data, through the configured coaxial polarized light source system (wavelength 630nm red LED + polarization angle adjustable filter), combined with a diffuser to suppress the specular reflection interference of metal / plastic materials, the imaging resolution is better than 50μm, and anti-reflection optimization is performed; Among them, in the module part of the high-resolution camera, a Peltier temperature control device is built-in, and the focal plane stability of ±0.1°C can still be maintained in the corrugated pipe production line environment of 40°C to avoid blurring of the corrugated pipe image data caused by thermal drift, and thermal stability is guaranteed; Multi-spectral light source: Develop a light compensation model based on the Retinex theory, use an adaptive light supplement algorithm to analyze the HSV color space histogram in real time, and dynamically adjust the light source intensity (adjustment range 50-1000Lux) to ensure that the detail retention rate in the dark area and the high-light area > 98%; Synchronization mechanism: An integrated incremental rotary encoder is adopted to obtain the production line speed of the corrugated pipe in real time (0 - 25 m / min). Synchronization trigger pulses are generated through the FPGA to ensure that the overlapping rate of adjacent frame corrugated pipe image data is ≥ 18%. The IEEE 1588 Precision Time Protocol is adopted to achieve a clock synchronization error of < 1 μs for multiple high-resolution cameras. Combined with the external trigger interface configuration of Halcon, the timing drift caused by network delay is eliminated. Spectral image processing: Convert the multi-spectral image into a single-channel grayscale image. Use median filtering to remove impulse noise and retain edge sharpness, as well as Gaussian filtering to smooth high-frequency random noise. Perform image enhancement and contrast stretching, and strengthen edge features and suppress low-frequency background interference through a high-pass filter. Perform perspective transformation or bilinear interpolation on the annular scanned corrugated pipe image to eliminate distortion caused by the shooting angle. Normalize the pixel values to the [0, 1] interval to improve the stability of model training. Equipment vibration data: Trigger the laser vibrometer through the synchronization mechanism, and use a dynamic signal acquisition card to collect the vibration signals generated during the operation of the corrugated pipe production line. Process the vibration signals to obtain equipment vibration data. Laser vibrometer: By debugging three non-contact laser vibrometers, there is no need for a coupling agent or mechanical contact with the corrugated pipe production line. Its accuracy can reach the micron level (displacement resolution of 15 pm) or even the nanometer level, and the velocity resolution is 0.01 μm / s (1 Hz). This avoids signal distortion caused by additional mass or poor contact of traditional sensors. Each laser vibrometer emits a laser beam with a fixed wavelength. The three laser beams at different angles are simultaneously focused on the measured point. Using the Doppler principle and combined with a three-dimensional vibration analysis algorithm, the internal and external vibration information during the production of the corrugated pipe production line is restored. Dynamic signal acquisition card: Convert the vibration information during the operation of the corrugated pipe production line into an analog electrical signal and input it to the analog front end of the acquisition card. The front-end circuit performs signal conditioning (such as amplification, filtering, and anti-aliasing processing), and then converts it into a digital signal through a high-precision ADC. All channels use a synchronous ADC clock to ensure phase consistency. Vibration signal preprocessing: Eliminate baseline drift, perform signal denoising and baseline correction. Baseline drift is mainly caused by sensor zero drift, such as mechanical vibration, and shows a low-frequency slow-varying trend (usually < 1 Hz). Use wavelet multi-resolution analysis to decompose the signal into sub-signals of different frequency bands, and subtract the low-frequency IMF component as the baseline. Filter out high-frequency noise higher than 1 / 2.56 of the sampling frequency through the anti-aliasing filter on the board of the dynamic signal acquisition card to prevent it from aliasing into the low-frequency band and causing distortion, and suppress high-frequency noise aliasing. For example, when the sampling rate is set to 10 kHz, the cut-off frequency is 3.9 kHz. Extract time-frequency domain features, perform data alignment and calibration, and extract the time-domain and frequency-domain features of the signal. The feature extraction module obtains the corrugated pipe image data, uses the synchronization mechanism to segment the corrugated pipe image data; based on the segmented results, extracts the head and tail image features and the pipe image features respectively as input features, obtains multiple groups of input features, and trains the distribution model; obtains the device vibration data and extracts the vibration features; among them, the distribution model includes a lightweight model and a high-precision model. Segment the corrugated pipe image data: Obtain the production line speed of the corrugated pipe, manually set the segmentation point of the corrugated pipe image data, map the segmentation point into the synchronization mechanism, and use the synchronization mechanism to segment the corrugated pipe image data: The synchronization mechanism includes time synchronization and hardware triggering; Through hardware triggering and time synchronization, use the timestamp alignment mechanism to achieve the segmentation of the corrugated pipe image data, ensuring the spatio-temporal consistency at the cutting points of the head, tail and the middle section data in the corrugated pipe image data; Hardware triggering: Use the corrugated pipe production line signal to synchronously trigger the high-resolution camera, perform a three-dimensional scanning of the corrugated pipe surface, detect the movement position and speed of the corrugated pipe at the end of the production line, and convert them into electrical signals or digital signals to ensure precise synchronization between devices; Time synchronization: Generate timestamps, record the time points when each synchronization mechanism is triggered, coordinate the simultaneous operation of multiple corrugated pipe production line devices, or align the working data between corrugated pipe production line devices through timestamps to ensure the consistency and accuracy of the data; Head and tail image features: Adopt the Canny algorithm to extract the edge contour of the corrugated pipe, separate the corrugated pipe area from the background area, and perform parameter verification; extract the geometric features of the head and tail ends, and perform texture feature verification and defect preliminary screening by deploying the MobileNetV3 model to extract the head and tail image features; Adopt the Canny algorithm to extract the edge contour of the corrugated pipe: Through the contour circumscribed circle fitting algorithm, calculate the diameter and fluctuation period of the corrugated pipe; through the statistics of the adjacent wave crest spacing, perform gradient calculation and Sobel operator convolution to obtain the horizontal and vertical three-dimensional matrix information of the corrugated pipe image data respectively, and obtain the Sobel operator calculation gradients and gradient amplitudes in the horizontal and vertical directions. The formula is: In the formula, I is the input corrugated pipe image data; * represents the convolution operation; is the Sobel operator calculation gradient in the horizontal direction; is the Sobel operator calculation gradient in the vertical direction.

[0019] According to the Sobel operator calculation gradients in the horizontal and vertical directions, calculate the gradient amplitude and gradient direction. The formula is: Wherein, is the amplitude; is the direction; the gradient direction is quantized into four main directions of 0°, 45°, 90°, and 135°. Along the gradient direction, the gradient amplitudes of the current bellows pixel and the two adjacent bellows pixels are compared. For example, if the gradient direction is 90°, the bellows pixels above and below are checked; when the gradient direction is not strictly aligned with the bellows pixel grid, the linear interpolation method is used to calculate the gradient values at non-integer position points to ensure the comparison accuracy. According to the characteristics of the bellows, the double thresholds can be set manually, or the double thresholds with typical parameters can also be used. Among them, the double thresholds include a high threshold and a low threshold; according to the edge connection rule, the bellows region and the background region are judged as follows: When the bellows pixel value is greater than the pixel high threshold, it is marked as a strong edge; When the bellows pixel value is greater than the pixel low threshold and less than or equal to the pixel high threshold, it is only retained when it is connected to the strong edge, otherwise it is suppressed; When the bellows pixel value is less than or equal to the pixel high threshold, it is marked as a weak edge and judged as the background image and discarded; To eliminate isolated noise points while retaining the periodicity and continuity of the bellows image data; Perform parameter verification: Align and verify the Canny geometric features to detect whether the bellows region and the background region are separated; Pipeline image features: By focusing on the recognition of the corrugation period and structural continuity of the bellows, the pipeline image features are extracted; For the longitudinal profile line of the pipeline, calculate the mean and standard deviation of the adjacent wave crest / trough spacing. The formula is: Wherein, is the mean corrugation period; is the standard deviation; is the adjacent wave crest / trough spacing; N is the total number of detected periods; i is the current period number, i = 1, 2, 3... N; Use the kurtosis index to evaluate the regularity of the corrugation shape, perform a fast Fourier transform, extract the main frequency component and its harmonic energy ratio, and quantify the consistency of the corrugation spacing. The formula is: Wherein, is the energy quantization value; is the spectrum amplitude; f is the spectrum amplitude input parameter; M is the total harmonic order; It is the main frequency energy ratio; n is the current harmonic order, where n = 1, 2, 3... M; when the main frequency energy accounts for a high proportion and the harmonics decay rapidly, it indicates excellent consistency of the corrugation spacing; for example, in the detection of corrugated tubes, when the main frequency energy accounts for more than 80%, it can be determined as a regular corrugated structure; Distribution model: Lightweight model call: Extract the features of the head and tail images, with geometric parameter verification as the core, call a low-complexity model, obtain the head and tail contours of the corrugated tube extracted by the Canny algorithm, calculate parameters such as diameter and corrugation spacing, compare with the standard template, and deploy a lightweight MobileNetV3 model to verify whether the geometric parameters exceed the tolerance. Its model parameter quantity is <1MB, and the single-frame inference delay is <5ms; High-precision model call: Extract the features of the middle pipeline image, decompose the image edge details and low-frequency background components using wavelet transform, extract texture features such as energy and contrast by combining the gray-level co-occurrence matrix, optimize the contrast of the uneven illumination area through the Retinex algorithm, strengthen the defect edges by combining frequency-domain high-pass filtering, adopt an improved YOLOv5s model, introduce the CBAM attention mechanism to enhance the focusing ability of the defect area, and improve the robustness of small-sample detection through hybrid data augmentation; Vibration characteristics: Extract vibration characteristics through multi-stage collaborative processing for characterization and dimensionality reduction. The specific process is as follows: S201: Time-frequency domain feature extraction: Time-domain statistics: Calculate kurtosis (impact sensitivity), peak factor, and impulse index (impact energy integral); Frequency-domain analysis: Extract the main frequency amplitude, spectral centroid, and frequency band energy ratio through FFT, and combine Hilbert envelope demodulation to detect the bearing fault characteristic frequency; Time-frequency combination: Use the db4 wavelet basis for 3-layer decomposition to calculate the node energy entropy, and combine short-time energy detection (50ms window, 75% overlap rate) to identify transient events (loosening, collision); S202: Nonlinear feature mining. Calculate the fractal dimension through the box-counting method to quantify the self-similarity and complexity of the signal, and combine the Lyapunov exponent to evaluate the chaotic characteristics of the system; S203: Feature optimization and selection: Redundancy elimination: Filter highly correlated features based on the Pearson correlation coefficient (threshold <0.8) and variance inflation factor (VIF <10); Importance ranking: Evaluate the feature contribution degree through the random forest algorithm, and retain the top-20 key features (such as kurtosis, rotational frequency harmonic energy); The anomaly judgment module judges the head and tail image anomalies and pipeline image anomalies in the input features according to the distribution model, and outputs the anomaly position area; obtains the vibration features, forms a vibration mapping with the anomaly position area, and establishes a vibration feedback; Segment the corrugated pipe image data according to the distribution model: After the distribution model identifies that the current corrugated pipe image data belongs to the head and tail ends or the middle end, the corresponding segmentation strategy is called: Head and tail ends: Use ellipse fitting or contour convex hull for geometric shape detection; Middle pipeline: Apply the watershed algorithm, combine with illumination compensation preprocessing or U-Net real-time segmentation network, and dynamically adjust the ROI area using the production line encoder signal to reduce the calculation amount; Head and tail image anomalies: Based on the Hough transform or YOLOv5-tiny lightweight detector, quickly locate the head and tail image areas to reduce redundant calculations; adopt the random cropping corrugated pipe image data cropping strategy of SAFE to retain the end face geometric features and avoid texture loss caused by downsampling, Weightedly fuse the geometric parameters (diameter, flatness) and texture features of the head and tail areas; The weight is dynamically adjusted by random forest; Perform TinyML quantization (INT8 precision) on the SAFE or UniVAD model, compress the model volume to within 500KB, and adapt to embedded devices such as ARM Cortex-M7; Pipeline image anomalies: Use YOLOv5-tiny as a fast positioning model for the head and tail areas (parameter quantity 1.7M, inference speed 15ms / frame), combined with the zero-shot anomaly detection ability of AnomalyGPT; divide cells along the pipeline axis, count the histogram of gradient directions in each cell, and capture local deformation features such as dents and bulges; calculate texture parameters such as contrast, energy, and homogeneity in the segmented area to identify defects such as surface oxidation and coating peeling; use deep learning feature fusion, input the corrugated pipe image data into the ViT model, extract the 768-dimensional embedding vector corresponding to the CLS token, fuse the global context information of the corrugations, and extract multi-scale feature maps through a U-Net-like network, combining shallow details and structural integrity features; Implement the "positioning - fine-grained analysis" process through dual-model cascading; AnomalyGPT uses a pre-trained visual encoder to extract image semantics, realizes threshold-free defect positioning through prompt engineering (such as "detect flange edge cracks"), introduces an LSTM network to process consecutive frame images, captures the dynamic change features of defects as the pipeline robot moves (such as crack propagation trend), and obtains the false detection rate caused by vibration in static images. The formula is: In the formula, is the waveform tube image anomaly index; is the geometric parameter flatness factor; Texture feature factor; is the edge blur weight; is the image stitching weight; is the harmonic order ripple spacing; is the harmonic attenuation rate; The anomaly grading mechanism is shown in Table 1: Table 1 Anomaly location area: By fusing the head and tail image anomalies and the temporal features of the pipeline image anomalies, a 3D convolutional network or Transformer is used to extract the temporal-spatial features, and the per-frame anomaly probability and the hotspot area mask are output; if the head and tail anomalies and the pipeline anomaly area are spatially continuous (such as the head anomaly extending towards the middle), it is confirmed as a real anomaly; Temporal feature fusion: Unify the head and tail image anomalies and the pipeline image anomalies to the same physical coordinate system; according to the production line speed v (which needs to be calibrated in real time by a laser rangefinder) and the shooting frame rate f, extract the temporal-spatial features of the physical position corresponding to the t-th frame of the temporal image; Enhance the texture features of the edge areas (bellows end faces) of the head and tail images using the Sobel operator, stack the temporal images into a 4D tensor of T×H×W×C (T = number of frames), and map them to the same dimension as F_pipe through a fully connected layer respectively; Head and tail anomaly detection: If the anomaly confidence of the head and tail images S_head > 0.9, record the anomaly area as the head end [0, x1]; Similarly, if S_tail > 0.9, record the tail end anomaly as [x2, L] (L is the pipe length); Pipeline anomaly propagation verification: Extract the X-axis range {[a_i, b_i]} of all anomaly areas in the pipeline temporal anomaly mask M_pipe; Continuity rule: Head extension: If there exists an a_i that satisfies |a_i - x1| < ε (ε = tolerance, such as 2mm), and the anomaly area in the subsequent frames continuously expands (a_{i+1} > a_i), then it is determined that the anomaly extends from the head end to the pipeline; Tail reverse extension: Similar rules are used to detect the anomaly diffusion from the tail to the front; Vibration mapping: At the detection points corresponding to the abnormal areas, compare the FRF curves of the experiment and the simulation, and analyze the amplitude difference and peak frequency shift; for example, when the abnormality causes a local stiffness decrease, the resonance peak frequency of the FRF will shift to a lower frequency and the amplitude will increase significantly; evaluate the weight of the vibration response of each node in the overall abnormality through the coordinate contribution amount, and locate the key areas sensitive to the abnormality; according to the vibration characteristics of the abnormal area (such as abnormal energy dissipation), adjust the local damping parameters of the simulation model to make the trend of the simulation results consistent with the test data; construct a transfer path model from the vibration excitation source to the abnormal area to quantify the attenuation law of the vibration energy in the structure; for example, change the transfer path through vibration isolation measures (active / passive vibration isolation) and verify its inhibitory effect on the vibration amplification in the abnormal area; Vibration feedback: Generate quantitative indicators such as the MAC matrix, FRAC (frequency response correlation) curve, etc., to visually display the matching degree between the vibration characteristics and the abnormal area; feedback the mapping result to the simulation model, and iteratively correct the material parameters, boundary conditions, etc., to improve the prediction accuracy of the model for real abnormalities; A real-time feedback module, based on vibration feedback, identifies the abnormal on-line forming of bellows; Obtain its dynamic characteristic parameters such as natural frequency and damping ratio; combined with Harmonic analysis, monitor the dynamic response of the bellows in the production line in real time, and accurately capture the forming problems caused by structural stress abnormalities; such as the performance indicators of the AI test system of Magao bellows, the experimental table is shown in Table 2: Table 2 Use the Kafka message queue to decouple the image acquisition and inference processes to ensure that single-frame end face detection is completed within 200 ms.

[0020] Example 2: Please refer to Figure 2 , based on Example 1, this example also provides a method for judging the abnormal on-line forming of bellows based on image recognition, including the following specific steps: Step 1: Obtain the bellows image data and the equipment vibration data; Step 2: Use the synchronization mechanism to segment the bellows image data; based on the segmented results, extract the input features, and the input features include the head and tail image features and the pipe image features; obtain multiple groups of input features and train the distribution model, and the distribution model includes a lightweight model and a high-precision model; based on the equipment vibration data, extract the vibration features; Step 3: According to the distribution model, judge the head and tail image abnormality and the pipe image abnormality in the input features, and output the abnormal position area; obtain the vibration mapping formed by the vibration features and the abnormal position area, and establish the vibration feedback; Step 4: Based on the vibration feedback, identify the anomalies in the on-line forming of the bellows.

[0021] The above embodiments can be implemented in whole or in part by software, hardware, firmware, or any combination thereof. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. Those of ordinary skill in the art will appreciate that the units and algorithm steps of the examples described in connection with the embodiments disclosed herein can be implemented in electronic hardware, or in a combination of computer software and electronic hardware. Whether these functions are executed in hardware or software depends on the specific application and design constraints of the technical solution.

[0022] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units. They may be located in one place or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0023] The above is only the specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present application, and all of them should be covered by the protection scope of the present application.

Claims

1. The system for judging abnormality of online bellows forming based on image recognition is characterized by: The system includes: Data acquisition preprocessing module, to obtain bellows image data and equipment vibration data; The feature extraction module uses a synchronization mechanism to segment the bellows image data; based on the segmentation results, the input features are extracted, including the head and tail image features and the pipe image features; multiple sets of input features are obtained to train distribution models, including lightweight models and high-precision models; vibration features are extracted based on the equipment vibration data; The abnormality judgment module judges the abnormality of the head and tail images and the pipeline image in the input features according to the distribution model, and outputs the abnormal location area; obtains the vibration mapping formed by the vibration feature and the abnormal location area, and establishes vibration feedback; The real-time feedback module identifies the anomalies of the online forming of the bellows based on the vibration feedback.

2. The system for judging abnormality of online forming of corrugated pipe based on image recognition according to claim 1 is characterized in that: The process of obtaining bellows image data and equipment vibration data is as follows: The synchronous mechanism is used to perform a three-dimensional annular scan on the surface of the bellows to obtain a multi-spectral image, and the spectral image is filtered to obtain the bellows image data; A dynamic signal acquisition card is used to collect vibration signals from the corrugated pipe production line during operation, and the baseline drift of the vibration signal is eliminated to obtain equipment vibration data.

3. The system for judging abnormality of online forming of corrugated pipe based on image recognition according to claim 1 is characterized in that: Segment the bellows image data: According to the speed of the bellows production line, the segmentation points of the bellows image data are set, the segmentation points are mapped to the synchronization mechanism, and the bellows image data is segmented using the synchronization mechanism.

4. The system for judging abnormality of online forming of corrugated pipe based on image recognition according to claim 3 is characterized in that: Extraction of first and last image features: The Canny algorithm is used to extract the edge contour of the bellows, and the horizontal and vertical three-dimensional matrix information of the bellows image data is obtained respectively. The Sobel operator in the horizontal and vertical directions is used to calculate the gradient and gradient amplitude. The formula is: Where I is the input bellows image data; Calculate the gradient for the Sobel operator in the horizontal direction; Calculate the gradient for the Sobel operator in the vertical direction; The gradient is calculated based on the Sobel operator in the horizontal and vertical directions, and the gradient amplitude and gradient direction are calculated. The formula is: In the formula, is the amplitude; For direction; Separate the bellows area from the background area and perform parameter verification; extract the geometric features of the head and tail ends, and deploy the MobileNetV3 model to verify the texture features and perform initial defect screening.

5. The system for judging abnormality of online forming of corrugated pipe based on image recognition according to claim 4 is characterized in that: Pipeline Image Features: The corrugation period and structural continuity of the bellows are identified through the bellows image data. Based on the corrugation period and structural continuity of the bellows, the mean and standard deviation of the distance between adjacent peaks and troughs are calculated. The formula is: In the formula, is the mean value of the ripple period; is the standard deviation; is the distance between adjacent peaks / troughs; N is the total number of detection cycles; i is the current cycle number, i=1,2,3...N; The kurtosis index is used to evaluate the regularity of the corrugation shape, and the fast Fourier transform is performed to extract the main frequency component and its harmonic energy ratio, and quantify the corrugation spacing. The formula is: In the formula, is the energy quantification value; is the spectrum amplitude; f is the spectrum amplitude input parameter; M is the total harmonic order; is the main frequency energy ratio; n is the current harmonic order, n=1,2,3...M; Vibration characteristics: Vibration features are extracted through multi-stage collaborative processing of vibration data characterization and dimensionality reduction.

6. The system for judging abnormality of online bellows forming based on image recognition according to claim 5 is characterized in that: Distributed model training: Lightweight model: Use the Canny algorithm to extract the contours of the bellows head and tail, and deploy a lightweight MobileNetV3 model to verify the surface geometric parameters of the bellows. High-precision model: Use wavelet transform to decompose the edge details and low-frequency background components of the bellows image data, optimize through the Retinex algorithm, obtain the contrast of the uneven lighting area, combine with the frequency domain high-pass filtering algorithm, strengthen the defect edge, and detect robustness.

7. The system for judging abnormality of online forming of corrugated pipe based on image recognition according to claim 6 is characterized in that: Head and tail image anomalies: Based on the YOLOv5-tiny lightweight detector, the head and tail image areas are located, and the SAFE random cropping strategy is used to crop the bellows image data, retain the end face geometric features, and perform weighted fusion of the geometric parameters of the head and tail areas with the texture features; Pipeline image anomalies: YOLOv5-tiny is used as a fast positioning model for the head and tail areas; cells are divided along the pipeline axis, and the gradient direction histogram in each cell is counted to capture the local deformation features of depressions and bulges; The LSTM network is introduced to process continuous frame images, capture the dynamic changing characteristics of defects along the pipeline, and obtain the false detection rate caused by vibration in static images.

8. The system for judging abnormality of online bellows forming based on image recognition according to claim 1 is characterized in that: Create vibration feedback: At the detection point corresponding to the abnormal area, the experimental and simulated FRF curves are compared to analyze the amplitude difference and peak frequency offset; the MAC matrix and FRAC curve quantitative indicators are generated to show the matching degree between the vibration characteristics and the abnormal area; The mapping results are fed back into the simulation model.

9. A method for judging abnormality of online bellows forming based on image recognition, using any system described in claims 1 to 8, characterized in that: The steps include: Step 1: Obtain bellows image data and equipment vibration data; Step 2: Use the synchronization mechanism to segment the bellows image data; extract input features based on the segmentation results, including head and tail image features and pipe image features; obtain multiple sets of input features and train distribution models, including lightweight models and high-precision models; extract vibration features based on equipment vibration data; Step 3: According to the distribution model, the head and tail image anomalies and the pipeline image anomalies in the input features are judged, and the abnormal position area is output; the vibration mapping formed by the vibration features and the abnormal position area is obtained, and vibration feedback is established; Step 4: Based on vibration feedback, identify the abnormality of bellows online forming.

Images