CN-122017178-A - Wire rope internal defect online detection system based on AI multisensor fusion

CN122017178ACN 122017178 ACN122017178 ACN 122017178ACN-122017178-A

Abstract

The invention discloses an AI multi-sensor fusion-based online detection system for internal defects of a steel wire rope, which relates to the technical field of defect detection and comprises a data preprocessing module, wherein the data preprocessing module synchronously acquires signal data streams of a plurality of heterogeneous sensors and performs preprocessing in the online operation process of the steel wire rope; the device comprises a feature decoupling module, an online detection module, a discriminator network, a wire rope internal defect online detection report, a causal characteristic analysis module and a wire rope internal defect online detection module, wherein the feature decoupling module is used for constructing a causal characteristic learning network, inputting a preprocessed signal data stream into the causal characteristic learning network, carrying out fusion and decoupling decomposition, generating a low-dimensional causal characteristic tensor and a high-dimensional confounding characteristic tensor, and calculating a decoupling loss value according to the separation degree of the low-dimensional causal characteristic tensor and the high-dimensional confounding characteristic tensor, and the online detection module is used for continuously collecting a new signal data stream, inputting the new signal data stream into the updated discriminator network, and outputting the online detection report of the wire rope internal defect. According to the invention, the causal characterization learning network is constructed, so that the characteristics of the fused multi-sensor are decomposed, and the accurate extraction of the essential characteristics of the defects and the effective isolation of interference signals are realized.

Inventors

CHEN JIAZHI
GU YUMU
CAO XIANG
CHEN JIANHAO
YANG JINYAN
LING CHEN
ZHOU CHUNXIU
CHU JING
YANG YILIN
MA XINYU
LIU XINGLONG
FANG HAOWEN
GONG JIAQI
Sheng Yinghao
JIA LIXIAO

Assignees

江苏航运职业技术学院

Dates

Publication Date: 20260512
Application Date: 20260122

Claims (10)

1. The wire rope internal defect online detection system based on AI multi-sensor fusion is characterized by comprising, The data preprocessing module synchronously collects signal data streams of a plurality of heterogeneous sensors and performs preprocessing in the online operation process of the steel wire rope; The feature decoupling module is used for constructing a causal characterization learning network, inputting the preprocessed signal data stream into the causal characterization learning network, carrying out fusion and decoupling decomposition, generating a low-dimensional causal feature tensor and a high-dimensional confounding feature tensor, and calculating a decoupling loss value according to the separation degree of the low-dimensional causal feature tensor and the high-dimensional confounding feature tensor; The countermeasure generation module combines the high-dimensional hybrid characteristic tensor with a preset random noise vector, inputs the combined high-dimensional hybrid characteristic tensor into a generator network and outputs a countermeasure characteristic sample; The parameter updating module inputs the antagonism characteristic sample and the low-dimensional causal characteristic tensor into the discriminator network and outputs a discrimination result difference; and the online detection module continuously collects new signal data flow, inputs the new signal data flow into the updated discriminator network and outputs an online detection report of the internal defects of the steel wire rope.
2. The wire rope internal defect online detection system based on AI multi-sensor fusion as recited in claim 1, wherein the plurality of heterogeneous sensors comprises a magnetic sensor, an acoustic sensor and an optical sensor; The signal data stream comprises magnetic flux leakage signal data, ultrasonic echo signal data and surface visual image data; The preprocessing comprises space-time registration, noise reduction filtering and amplitude normalization.
3. The online detection system for defects in a steel wire rope based on AI multi-sensor fusion as set forth in claim 2, wherein said constructing a causal characterization learning network comprises the steps of, Building a time sequence feature coding layer by using a one-dimensional convolutional neural network, building a space feature coding layer by using a two-dimensional convolutional neural network, and building a feature fusion and decoupling layer by using a cross-modal attention mechanism; And carrying out parallel processing and fusion on the time sequence feature coding layer, the space feature coding layer and the feature fusion and decoupling layer to generate a causal characterization learning network.
4. The wire rope internal defect online detection system based on AI multisensor fusion of claim 3, wherein the generating of the low-dimensional causal feature tensor and the high-dimensional confounding feature tensor comprises the steps of, Inputting the preprocessed signal data stream into a time sequence feature coding layer and a space feature coding layer, and extracting time sequence features and space features; Performing cross-modal attention fusion on the time sequence feature and space feature input feature fusion and decoupling layer to generate a primary fusion feature; Based on the primary fusion features, calculating importance scores of feature dimensions in the primary fusion features, and weighting the primary fusion features according to the importance scores to obtain causal enhancement features; and inputting the causal enhancement features into a decoupling encoder, and respectively performing dimension mapping and regularization constraint through two parallel full-connection layers to generate a low-dimensional causal feature tensor and a high-dimensional confounding feature tensor.
5. The online detection system for defects in a steel wire rope based on AI multi-sensor fusion as set forth in claim 4, wherein said calculating a decoupling loss value comprises the steps of, Calculating a global correlation measure between the low-dimensional causal feature tensor and the high-dimensional confounding feature tensor by a dynamic weighting multi-scale mutual information estimator; invoking a historical online detection result, and evaluating causal driving strength of the low-dimensional causal feature tensor based on a mapping relation between the historical online detection result and the low-dimensional causal feature tensor; And carrying out weighted fusion on the global correlation measure and the causal driving strength to generate a decoupling loss value.
6. The wire rope internal defect online detection system based on AI multisensor fusion of claim 5, wherein the outputting of the resistance characteristic sample comprises the steps of, Extracting an energy distribution feature vector by carrying out multi-scale spectrum analysis on the high-dimensional hybrid feature tensor, and mapping a preset random noise vector into a condition vector through a condition coding network; Generating a dynamic weight matrix based on the energy distribution feature vector and the condition vector; inputting the high-dimensional hybrid feature tensor and a preset random noise vector into a trunk encoder of a generator network to obtain an initial feature map; And performing conditional spectrum modulation on the initial feature map by using a dynamic weight matrix to generate a modulated feature map, inputting the modulated feature map into a decoder of a generator network, and outputting an antagonistic feature sample.
7. The online detection system for defects inside a steel wire rope based on AI multi-sensor fusion as set forth in claim 6, wherein said outputting a difference in discrimination results comprises the steps of, Feature fusion is carried out on the antagonism feature sample and the low-dimensional causal feature tensor through an interaction attention mechanism built in the discriminator network, so that interaction enhancement features are generated; Dynamically selecting a processing path according to local information entropy and global variance of the interaction enhancement features, performing differentiation processing on detail features and context features in the interaction enhancement features, and generating path modulation features; And extracting multi-scale gradient statistical features of the path modulation features, performing attention weighted multivariable nonlinear mapping on the multi-scale gradient statistical features, and generating discrimination result differences.
8. The online detection system for defects in a steel wire rope based on AI multi-sensor fusion as set forth in claim 7, wherein said calculating an antagonism loss value based on the difference of the discrimination results comprises the steps of, Performing multi-scale time sequence decomposition on the discrimination result difference, and extracting difference components on different time scales; Calculating a weighting coefficient based on the variance of each difference component and the covariance of each difference component and the difference of the discrimination result; and integrating and fusing the difference components by using the weighting coefficients to generate an antagonism loss value.
9. The wire rope internal defect online detection system based on AI multisensor fusion of claim 8, wherein the passing through the resistive loss value and the decoupling loss value updates parameters of the generator network and the discriminator network together, as follows, Inputting the antagonism loss value and the decoupling loss value into a dual-channel weight distribution network, extracting dynamic characteristics of each loss value, and generating dynamic fusion weights by using a cross attention mechanism; and weighting and fusing the antagonism loss value and the decoupling loss value into a total loss function based on the dynamic fusion weight, and synchronously updating parameters of the generator network and the discriminator network through a gradient descent algorithm.
10. The online detection system for the internal defects of the steel wire rope based on AI multi-sensor fusion as set forth in claim 9, wherein the online detection report of the internal defects of the steel wire rope is output by the steps of, Inputting the continuously collected new signal data stream into the updated discriminator network, outputting a defect detection result, and performing multi-scale time sequence verification on the defect detection result to generate a real-time confidence evaluation result; Synthesizing a real-time confidence assessment result based on a multi-source evidence fusion theory to generate comprehensive confidence distribution; and carrying out multi-mode information fusion processing on the comprehensive confidence distribution and the low-dimensional causal feature tensor through a joint decision mechanism to generate a structured detection report.

Description

Wire rope internal defect online detection system based on AI multisensor fusion Technical Field The invention relates to the technical field of defect detection, in particular to an online detection system for internal defects of a steel wire rope based on AI multi-sensor fusion. Background In recent years, a steel wire rope nondestructive testing technology based on multi-sensor fusion gradually becomes a research hot spot in the field of industrial safety monitoring. Complementary information of the internal state of the steel wire rope can be obtained through integrating magnetic, acoustic, optical and other heterogeneous sensors, and the automatic extraction and classification of defect characteristics are realized by combining a deep learning algorithm. In the prior art, a convolutional neural network or a cyclic neural network is mostly adopted to perform feature level or decision level fusion on multi-source data, so that the accuracy and the robustness of defect identification are improved, and in addition, the generation of an countermeasure network is introduced to enhance the adaptability of a model to complex working conditions. However, the existing method generally lacks explicit modeling on causal relation in the feature fusion process, so that the extracted features are easily interfered by non-defect related factors such as environmental noise and load fluctuation, and the clutter factors are highly coupled with the real defect features on the data level, so that the models are difficult to distinguish the related features of the intrinsic causal features and the appearance, and particularly under the on-line detection scene, the problem of feature confusion is further amplified due to the dynamically-changed working condition, and the generalization capability and reliability of the detection models are restricted. Disclosure of Invention The present invention has been made in view of the above-described problems occurring in the prior art. Therefore, the invention provides an AI multi-sensor fusion-based online detection system for the defects in the steel wire rope, which solves the problems of insufficient generalization capability and high false alarm rate caused by feature confusion in the existing detection method for the defects in the steel wire rope. In order to solve the technical problems, the invention provides the following technical scheme: The invention provides an AI multi-sensor fusion-based steel wire rope internal defect online detection system which comprises a data preprocessing module, a characteristic decoupling module, an online detection module, an antagonism generation module, a parameter updating module, a parameter reporting module and an online detection module, wherein the data preprocessing module synchronously acquires signal data streams of a plurality of heterogeneous sensors in the online operation process of the steel wire rope, performs preprocessing, constructs a causality characterization learning network, inputs the preprocessed signal data streams into the causality characterization learning network, performs fusion and decoupling decomposition, generates a low-dimensional causality tensor and a high-dimensional confounding characteristic tensor, calculates a decoupling loss value according to the separation degree of the low-dimensional causality tensor and the high-dimensional confounding characteristic tensor, combines the high-dimensional confounding characteristic tensor with a preset random noise vector, inputs the high-dimensional confounding characteristic tensor into the generator network, outputs an antagonism characteristic sample, inputs the antagonism characteristic sample into the low-dimensional causality tensor, outputs a discrimination result difference, calculates an antagonism loss value based on the discrimination result difference, updates parameters of the generator network and the discriminator network together through the antagonism loss value and the decoupling loss value, continuously acquires a new signal data stream, and outputs the updated signal data stream to the online detector to report the online defect. As an optimal scheme of the online detection system for the internal defects of the steel wire rope based on AI multi-sensor fusion, the invention comprises a plurality of heterogeneous sensors, a detection system and a detection system, wherein the heterogeneous sensors comprise a magnetic sensor, an acoustic sensor and an optical sensor; The signal data stream comprises magnetic flux leakage signal data, ultrasonic echo signal data and surface visual image data; The preprocessing comprises space-time registration, noise reduction filtering and amplitude normalization. As an optimal scheme of the online detection system for the internal defects of the steel wire rope based on AI multi-sensor fusion, the method comprises the following steps of constructing a causal characterization learning network, Building a time