Intelligent AI identification method based on multi-sensor data fusion
Through multi-sensor data fusion and deep neural network technology, the problem of insufficient accuracy and classification capabilities in bridge damage recognition is solved, and higher recognition accuracy and data security are achieved, and the accuracy and system of bridge structure health assessment are improved. stability.
Patent Information
- Application Number
- CN202411281506.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-13
- Publication Date
- 2024-12-06
- Estimated Expiration
- 2044-09-13
AI Technical Summary
The existing bridge damage recognition methods have problems such as poor accuracy, inability to obtain all-round information, weak classification capabilities, and low recognition accuracy, especially due to the subjectivity of manual visual evaluation and the difficulty of traditional methods to handle the nonlinear relationship of multi-sensor data. .
Intelligent AI recognition method using multi-sensor data fusion, including installing multiple sensors on the bridge, transmitting data wirelessly to the central processing center through encrypting, filtering and cleaning data using an adaptive learning module, and applying wavelet threshold decomposition and depth based on parameter optimization Neural network technology performs data analysis, predicts bridge damage and maintenance needs, and uses fault tolerance mechanisms and encryption protocols to ensure data security and integrity.
Improve the accuracy and prediction accuracy of bridge damage identification, enhance the comprehensiveness and safety of data, improve the accuracy of bridge structure health assessment and system stability, increase the accuracy of bridge current status assessment by 30%, predict potential The accuracy of damage and maintenance requirements is increased by 35%.
Smart Images

Figure CN119089124A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of intelligent AI recognition methods, and specifically to an intelligent AI recognition method based on multi-sensor data fusion. Background Art
[0002] When evaluating bridge structures, more and more bridges are equipped with structural health monitoring systems. These monitoring systems continuously collect various sensor data of bridges, including dynamic response, static response and apparent morphology. These data containing a large amount of damage information are the basis for bridge status assessment. Therefore, how to interpret these data from the perspective of structural safety has become the focus of bridge damage identification research. With the development of society, the monitoring accuracy requirements for bridge structures are getting higher and higher. It is no longer the era of monitoring with only a single monitoring device. Researchers use multiple types or multiple monitoring devices of the same type to measure the bridge in multiple dimensions at the same time. Only in this way can more complete information about the bridge structure be collected in the laboratory for in-depth research. The most commonly used method for bridge damage identification today is manual visual inspection. In a typical bridge inspection, trained inspectors closely inspect the various components of the bridge structure. They evaluate the condition of bridge components and give them ratings. This rating is based on a set of guidelines and the inspector's experience and subjective evaluation of the current situation. For most cases, this type of evaluation is appropriate and effective.
[0003] Manual visual assessment methods are highly subjective, and the raw data of bridge condition assessments are subject to noise interference. Therefore, the technical level and experience of the inspector have an important impact on the accuracy of bridge damage diagnosis, and this impact is often uncontrollable. The lack of obvious feature information cannot effectively fuse and analyze the data of multiple sensors, and it is impossible to obtain comprehensive information on damage and disease. Traditional recognition methods find it difficult to model the complex nonlinear relationship between data and bridge structural damage, so it is difficult to reflect the corresponding relationship between the implicit information of the data and the structural damage, resulting in weak classification ability and low recognition accuracy in practical applications. Summary of the invention
[0004] 1. Technical issues to be resolved
[0005] In view of the shortcomings of the prior art, the present invention provides an intelligent AI recognition method based on multi-sensor data fusion, which solves the problems of poor accuracy, inability to obtain comprehensive information on injuries and diseases, weak classification ability and low recognition accuracy in practical applications.
[0006] (II) Technical solution
[0007] To achieve the above objectives, the present invention is implemented through the following technical solutions: 1. An intelligent AI recognition method based on multi-sensor data fusion, characterized in that it includes:
[0008] S1. Install various types of sensors on the bridge structure, such as accelerometers, strain gauges, and temperature sensors, to collect real-time operating data of the bridge and send it to the central data processing center through encrypted wireless transmission. This step emphasizes the advantages of multi-sensor data fusion, and improves the comprehensiveness and accuracy of the data through the complementary data of different types of sensors;
[0009] S2. In the central data processing center, the adaptive learning module is used to preliminarily screen and clean the received data to filter out noise and erroneous data and improve data quality. The adaptive learning module continuously learns and adjusts the screening rules to adapt to changes in different environments and data sources, thereby improving the flexibility and effectiveness of data processing;
[0010] S3. Further process the data using a wavelet threshold decomposition method based on parameter optimization. This method adjusts the wavelet coefficients according to the provided function expression to optimize the data quality.
[0011]
[0012] Among them, Wj,k is the wavelet coefficient, λ is the threshold, α and β are the adjustment factors, si gn is the sign function, and after parameter adjustment and algorithm optimization, the optimal adjustment factor and the improved threshold function eliminate the discontinuity at the threshold and make the function quickly approach the hard threshold function;
[0013] S4. Apply machine learning algorithms using deep neural network technology to conduct in-depth analysis of the processed data to assess the current status and structural health of the bridge;
[0014] S5. The machine learning algorithm self-adjusts its parameters based on the equipment operation data to adapt to the aging process of the bridge and the changing environmental conditions;
[0015] S6. Predict potential bridge damage and maintenance needs based on the output of the machine learning algorithm, and update the damage prediction model to integrate the latest bridge performance data and maintenance feedback;
[0016] S7, automatically sending the prediction results and maintenance suggestions to the operation and maintenance manager and displaying them through the user interface, allowing the operation and maintenance manager to view the bridge status, prediction results and execute remote troubleshooting instructions;
[0017] S8. Adopt data encryption and verification mechanisms, use advanced encryption standards to ensure the security and integrity of all encrypted wireless transmission and storage data, and implement data access rights verification through multi-factor authentication to prevent data leakage and unauthorized access;
[0018] S9. Use fault-tolerance mechanisms, including local data caching and automatic data recovery, to ensure that the system continues to operate and maintain the highest possible data integrity in the event of unstable network connections or partial data loss;
[0019] S10, the data processing and analysis module is equipped with self-diagnosis to detect and report system internal errors and performance bottlenecks.
[0020] Preferably, the wavelet threshold decomposition method uses Daubechies wavelet to perform four-layer wavelet decomposition, and adopts a threshold function between a hard threshold and a soft threshold to eliminate noise.
[0021] Preferably, the machine learning algorithm includes a fast Fourier transform for extracting frequency domain features from time series data, and uses a principal component analysis method to reduce the dimension of the features.
[0022] Preferably, the prediction model adopts an online learning method, which can integrate new data feedback in real time to improve the accuracy of the prediction and the adaptability of the model. The user interface has touch and voice control functions to facilitate managers to operate more conveniently in complex environments.
[0023] Preferably, the encryption protocol includes but is not limited to TLS / SSL to ensure end-to-end security during data transmission, and the data recovery function uses a cloud-based backup system to asynchronously replicate data at multiple locations to ensure high data availability and disaster recovery capabilities.
[0024] Preferably, the self-diagnosis function includes applying machine learning technology to analyze system operation logs to automatically identify potential system anomalies and performance degradation trends, and the parameter self-adjustment mechanism optimizes its identification algorithm and parameter settings based on environmental changes and historical maintenance data.
[0025] Preferably, the encrypted wireless transmission is digitally signed and integrity checked before transmission to prevent data tampering during transmission, and the deep neural network technology includes a combination of convolutional neural networks and recursive neural networks to process and analyze timing-dependent data characteristics.
[0026] Preferably, the data is cleaned, and the data cleaning process includes using an outlier detection algorithm to automatically identify and remove statistically abnormal data points. The data denoising process uses an adaptive filtering technology based on machine learning to improve the accuracy and efficiency of data cleaning.
[0027] (III) Beneficial effects
[0028] The present invention provides an intelligent AI recognition method based on multi-sensor data fusion.
[0029] Beneficial effects:
[0030] The present invention proposes a wavelet threshold decomposition based on parameter optimization. The method improves and optimizes the traditional wavelet threshold function. Two adjustment parameters and are introduced through a random optimization algorithm to adapt to different wavelet decomposition layers, thereby further removing noise signals. A feature engineering method based on sensor data mining is proposed, which uses fast Fourier transform and sliding window to extract modal frequencies. The latter uses principal component analysis to fuse bridge damage information and reduce feature dimensions of monitoring data. The main function of this feature engineering is to transform the original signal from the time domain into analysis in the frequency domain through fast Fourier transform, and use a sliding window to extract modal frequency features from the spectrum diagram, fully mining hidden information sensitive to bridge damage, and finally using the sensitive features with large damage information selected by principal component analysis as input for subsequent bridge damage identification. BRIEF DESCRIPTION OF THE DRAWINGS
[0031] Figure 1 This is the technical solution of the present invention.
[0032] Figure 2 For the proposed feature engineering.
[0033] Figure 3 Schematic diagram of wavelet decomposition.
[0034] Figure 4 is the wavelet threshold decomposition.
[0035] Figure 5 The principal component analysis method is used.
[0036] Figure 6 Schematic diagram of sliding window.
[0037] Figure 7 This is the damage identification process of the present invention.
[0038] Figure 8 A schematic diagram of a neural network.
[0039] Fig. 9 Schematic diagram of transfer learning. DETAILED DESCRIPTION
[0040] The technical solutions in the embodiments of the present invention are described clearly and completely. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0041] An embodiment of the present invention provides an intelligent AI recognition method based on multi-sensor data fusion, including: S1, installing various types of sensors on the bridge structure, such as accelerometers, strain gauges, and temperature sensors, for real-time collection of bridge operation data, and sending the data to a central data processing center via encrypted wireless transmission. This step emphasizes the advantages of multi-sensor data fusion and improves the comprehensiveness and accuracy of the data through data complementarity of different types of sensors.
[0042] S2. In the central data processing center, the adaptive learning module is used to perform preliminary screening and cleaning of the received data to filter out noise and erroneous data and improve data quality. The adaptive learning module continuously learns and adjusts the screening rules to adapt to changes in different environments and data sources, thereby improving the flexibility and effectiveness of data processing.
[0043] S3. Further process the data using a wavelet threshold decomposition method based on parameter optimization. This method adjusts the wavelet coefficients according to the provided function expression to optimize the data quality.
[0044]
[0045] Among them, Wj,k is the wavelet coefficient, λ is the threshold, α and β are adjustment factors, si gn is the sign function. After parameter adjustment and algorithm optimization, the optimal adjustment factor, the improved threshold function eliminates the discontinuity at the threshold and makes the function quickly approach the hard threshold function. The wavelet threshold decomposition method uses Daubech i es wavelet for four-layer wavelet decomposition, and uses a threshold function between the hard threshold and the soft threshold for noise elimination.
[0046] S4. Apply machine learning algorithms with deep neural network technology to conduct in-depth analysis of the processed data to evaluate the current status and structural health of the bridge. The machine learning algorithm includes fast Fourier transform to extract frequency domain features from time series data, and use principal component analysis to reduce the dimension of the features. Fast Fourier transform is used to analyze the characteristic laws of the original data. The signal is transformed into a function with frequency as the independent variable by Fourier transform, and the original time series data is transformed from time domain analysis to frequency domain analysis. When the bridge structure is healthy, the vibration signal on its spectrum is usually a low-frequency signal. When the bridge is damaged, high-frequency signals are obviously seen in the signal. The frequency characteristics of the bridge are changed. The spectrum information obtained from the frequency domain shows the dependence difference of the damage category that is not shown in the time domain. Therefore, the present invention selects the modal frequency as the feature of damage identification. Specifically, all data features are sampled over the entire bridge monitoring data length, and the sliding window size is selected to be twice the data sampling frequency, with a step size of 1. The modal frequency characteristics of the maximum amplitude point and the second largest amplitude point in the spectrum diagram are extracted for each sampling point in each sliding window. The input signal size is the same during each fast Fourier transform. Finally, the output result is normalized, and the processed data has better stability than the original monitoring data.
[0047] S5. The machine learning algorithm self-adjusts its parameters based on equipment operation data to adapt to the aging process of the bridge and changing environmental conditions.
[0048] S6. Based on the output of the machine learning algorithm, the potential damage and maintenance needs of the bridge are predicted, and the damage prediction model is updated to integrate the latest bridge performance data and maintenance feedback. The prediction model adopts online learning, which can integrate new data feedback in real time, improve the accuracy of the prediction and the adaptability of the model. The user interface has touch and voice control functions to facilitate managers to operate more conveniently in complex environments. Through deep neural network technology and self-adjustment of machine learning algorithms, the accuracy of bridge health assessment has been significantly improved. The specific evaluation results are as follows: the accuracy of the assessment of the current state and structural health of the bridge has increased by about 30%, and the accuracy of predicting potential damage and maintenance needs has increased by about 35%.
[0049] S7. The prediction results and maintenance recommendations are automatically sent to the operation and maintenance managers and displayed through the user interface, allowing the operation and maintenance managers to view the bridge status, prediction results, and execute remote troubleshooting instructions.
[0050] S8. Adopt data encryption and verification mechanisms, use advanced encryption standards to ensure the security and integrity of all encrypted wirelessly transmitted and stored data, and implement data access permission verification through multi-factor authentication to prevent data leakage and unauthorized access. Encryption protocols include but are not limited to TLS / SSL to ensure end-to-end security during data transmission. The data recovery function uses a cloud-based backup system and asynchronously replicates data at multiple locations to ensure high data availability and disaster recovery capabilities. Encrypted wireless transmission is digitally signed and integrity checked before transmission to prevent data tampering during transmission. Deep neural network technology includes a combination of convolutional neural networks and recurrent neural networks to process and analyze time-dependent data characteristics.
[0051] S9. Adopt fault-tolerant mechanisms, including local data caching and automatic data recovery functions, to ensure that the system can continue to run and maintain the highest possible data integrity when the network connection is unstable or some data is lost. Local data caching: The system will store a copy of the data locally. When there is a network connection problem or data is lost during data transmission, the system can obtain data from the local cache to ensure data availability. This caching mechanism can reduce the reliance on real-time data transmission and improve the stability of the system. Automatic data recovery function: The system will monitor abnormal conditions during data transmission. When data loss or corruption is detected, the data recovery mechanism will be automatically triggered, which may include re-requesting lost data blocks or using redundant data for repair. Through the automated data recovery function, the system can maintain data integrity as much as possible and ensure data accuracy and reliability.
[0052] S10. The data processing and analysis module is equipped with self-diagnosis to detect and report system internal errors and performance bottlenecks. The self-diagnosis function includes applying machine learning technology to analyze the system operation log, automatically identifying potential system anomalies and performance degradation trends, and the parameter self-adjustment mechanism optimizes its recognition algorithm and parameter settings according to environmental changes and historical maintenance data. The data is cleaned. The data cleaning process includes using an outlier detection algorithm to automatically identify and remove statistically abnormal data points. The data denoising process uses adaptive filtering technology based on machine learning to improve the accuracy and efficiency of data cleaning. Data encryption transmission delay: 0.05 seconds, data recovery time under the fault-tolerant mechanism: 2 seconds, self-diagnosis and error reporting accuracy: 99%.
[0053] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.
Claims
1. An intelligent AI recognition method based on multi-sensor data fusion, characterized in that: include: S1. Install various types of sensors on the bridge structure, such as accelerometers, strain gauges, and temperature sensors, to collect real-time operating data of the bridge and send it to the central data processing center through encrypted wireless transmission. This step emphasizes the advantages of multi-sensor data fusion, and improves the comprehensiveness and accuracy of the data through the complementary data of different types of sensors; S2. In the central data processing center, the adaptive learning module is used to preliminarily screen and clean the received data to filter out noise and erroneous data and improve data quality. The adaptive learning module continuously learns and adjusts the screening rules to adapt to changes in different environments and data sources, thereby improving the flexibility and effectiveness of data processing; S3. Further process the data using a wavelet threshold decomposition method based on parameter optimization. This method adjusts the wavelet coefficients according to the provided function expression to optimize the data quality. Among them, Wj,k is the wavelet coefficient, λ is the threshold, α and β are the adjustment factors, and sign is the sign function. After parameter adjustment and algorithm optimization, the optimal adjustment factor and the improved threshold function eliminate the discontinuity at the threshold and make the function quickly approach the hard threshold function; S4. Apply machine learning algorithms using deep neural network technology to conduct in-depth analysis of the processed data to assess the current status and structural health of the bridge; S5. The machine learning algorithm self-adjusts its parameters based on the equipment operation data to adapt to the aging process of the bridge and the changing environmental conditions; S6. Predict potential bridge damage and maintenance needs based on the output of the machine learning algorithm, and update the damage prediction model to integrate the latest bridge performance data and maintenance feedback; S7, automatically sending the prediction results and maintenance suggestions to the operation and maintenance manager and displaying them through the user interface, allowing the operation and maintenance manager to view the bridge status, prediction results and execute remote troubleshooting instructions; S8. Adopt data encryption and verification mechanisms, use advanced encryption standards to ensure the security and integrity of all encrypted wireless transmission and storage data, and implement data access rights verification through multi-factor authentication to prevent data leakage and unauthorized access; S9. Use fault-tolerance mechanisms, including local data caching and automatic data recovery, to ensure that the system continues to operate and maintain the highest possible data integrity in the event of unstable network connections or partial data loss; S10, the data processing and analysis module is equipped with self-diagnosis to detect and report system internal errors and performance bottlenecks.
2. The intelligent AI recognition method based on multi-sensor data fusion according to claim 1, characterized in that: The wavelet threshold decomposition method uses Daubechies wavelet to perform four-layer wavelet decomposition, and adopts a threshold function between a hard threshold and a soft threshold to eliminate noise.
3. The intelligent AI recognition method based on multi-sensor data fusion according to claim 1, characterized in that: The machine learning algorithm includes fast Fourier transform for extracting frequency domain features from time series data, and uses principal component analysis method to reduce the dimension of the features.
4. The intelligent AI recognition method based on multi-sensor data fusion according to claim 1, characterized in that: The prediction model adopts an online learning method and can integrate new data feedback in real time to improve the accuracy of the prediction and the adaptability of the model. The user interface has touch and voice control functions to facilitate managers to operate more conveniently in complex environments.
5. The intelligent AI recognition method based on multi-sensor data fusion according to claim 1, characterized in that: The encryption protocols include but are not limited to TLS / SSL, which ensure end-to-end security during data transmission. The data recovery function uses a cloud-based backup system to asynchronously replicate data at multiple locations to ensure high data availability and disaster recovery capabilities.
6. The intelligent AI recognition method based on multi-sensor data fusion according to claim 1, characterized in that: The self-diagnosis function includes applying machine learning technology to analyze system operation logs to automatically identify potential system anomalies and performance degradation trends. The parameter self-adjustment mechanism optimizes its identification algorithm and parameter settings based on environmental changes and historical maintenance data.
7. The intelligent AI recognition method based on multi-sensor data fusion according to claim 1, characterized in that: The encrypted wireless transmission is digitally signed and integrity checked before transmission to prevent data from being tampered with during transmission. The deep neural network technology includes a combination of convolutional neural networks and recursive neural networks to process and analyze timing-dependent data characteristics.
8. The intelligent AI recognition method based on multi-sensor data fusion according to claim 1, characterized in that: The data is cleaned using data cleaning, and the data cleaning process includes using an outlier detection algorithm to automatically identify and remove statistically abnormal data points. The data denoising process uses an adaptive filtering technology based on machine learning to improve the accuracy and efficiency of data cleaning.
Citation Information
Patent Citations
Method for extracting Beidou bridge deformation monitoring dynamic characteristics based on wavelet analysis
CN112762886A
Monitoring signal filtering method based on wavelet analysis and threshold processing
CN113405747A
Wavelet transform algorithm micro-magnetic detector bridge cable damage detection method
CN115494147A
Bridge modal parameter identification method based on array mobile sensor
CN116756611A
Intelligent bridge detection method and system based on multi-source detection fusion
CN117782212A
