Engineering vibration detection system based on machine vision

Abstract

The invention relates to the technical field of vibration detection, in particular to an engineering vibration detection system based on machine vision, which comprises a high-speed image capture module, a vibration feature extraction module, a frequency response mapping module and a data fusion detection module. According to the method, the video data of the surface of the engineering structure are continuously captured, the time sequence data are extracted from the video data, high-precision monitoring of fine vibration fluctuation is achieved, the key vibration frequency and amplitude can be recognized more accurately through time-frequency analysis, and by constructing the mapping relation between the frequency and the amplitude, the accuracy of the vibration fluctuation is improved. The system effectively reveals the subtle change of the vibration frequency, can carry out predictive maintenance and early fault recognition, enhances the reliability of data and the depth of analysis, comprehensively analyzes and recognizes deviation and abnormal modes, greatly improves the accuracy and timeliness of structural health monitoring, and remarkably reduces the maintenance cost and downtime.

Description

Technical Field

[0001] The invention relates to the technical field of vibration detection, and in particular to an engineering vibration detection system based on machine vision. Background Art

[0002] The field of vibration detection technology mainly involves the use of various sensors and devices to monitor and analyze the vibrations generated by mechanical equipment, structures and systems during operation. It can identify abnormal vibrations and help maintenance personnel diagnose potential failures or damage. Vibration analysis can evaluate the health of equipment and determine whether the equipment is within the normal operating range through frequency analysis, amplitude evaluation and waveform comparison. In addition, vibration detection is also widely used in predictive maintenance strategies to preventively identify potential failures through continuous monitoring and trend analysis, thereby reducing downtime and maintenance costs.

[0003] Among them, the engineering vibration detection system is an application of vibration detection technology, which is mainly used to monitor the vibration conditions of building structures, bridges, industrial equipment, etc. It can capture data in real time and evaluate the stability of the structure and the operating efficiency of the equipment by analyzing the characteristics of vibration. The purpose is to ensure structural safety, optimize mechanical performance, and provide early warning of potential structural failure risks, thereby ensuring personnel safety and long-term operation of equipment.

[0004] Existing vibration detection technologies are physically limited in data collection coverage and frequency resolution, making it difficult to capture tiny or high-frequency vibrations. Frequency analysis and waveform comparison in traditional technologies rely on post-data processing and cannot provide real-time feedback. In engineering applications, this may result in an untimely response to structural problems, increasing safety risks. Due to the single data source, the accuracy and reliability of traditional methods in complex or interfering environments are often affected. The lack of real-time data analysis and comprehensive trend prediction also limits the application effect of predictive maintenance strategies, and data-driven decision support cannot be fully utilized, resulting in greater economic losses and safety accidents due to failure to predict and handle potential failures in a timely manner. Summary of the invention

[0005] The purpose of the present invention is to solve the shortcomings in the prior art and to propose an engineering vibration detection system based on machine vision.

[0006] In order to achieve the above object, the present invention adopts the following technical solution: an engineering vibration detection system based on machine vision comprises:

[0007] The high-speed image capture module continuously records video data of the surface of the engineering structure through a camera, captures time series data from the video stream, and generates vibration fluctuation signal data;

[0008] The vibration feature extraction module performs time-frequency analysis on the signal based on the vibration fluctuation signal data, identifies the key vibration frequency, calculates the amplitude according to the key vibration frequency identification result, extracts and classifies the difference vibration mode, and obtains the vibration feature data set;

[0009] The frequency response mapping module calculates the frequency value of each data point based on the vibration characteristic data set, constructs a mapping relationship between frequency and amplitude, reveals the frequency change according to the mapping relationship, corresponds to the amplitude, and updates and draws a frequency response graph in real time;

[0010] The data fusion detection module fuses the frequency response diagram with the vibration sensor data, performs data comparison and historical trend analysis, identifies deviations based on the analysis results, determines abnormal vibration modes, and generates engineering structure health analysis results.

[0011] As a further solution of the present invention, the step of acquiring the vibration fluctuation signal data is:

[0012] Install cameras at key locations of engineering structures to capture surface changes in real time and generate real-time video stream data;

[0013] Performing image analysis on the real-time video stream data, identifying motion information and changes in each frame, extracting time series, and obtaining a time series data set;

[0014] Based on the time series data set, the formula is adopted:

[0015]

[0016] Calculate the change Δs between frames to obtain the vibration fluctuation signal data, where frame i and frame i-1 They represent the i-th and i-1-th key frames respectively, and m represents the total number of key frames.

[0017] As a further solution of the present invention, the step of identifying the key vibration frequency is:

[0018] Performing time-frequency conversion on the vibration fluctuation signal data, analyzing the frequency components of the signal at the difference time points by short-time Fourier transform, and generating a signal time-frequency diagram;

[0019] According to the signal time-frequency diagram, peak detection is performed to identify the frequency points where key energy is concentrated, and a frequency candidate list is generated; based on the frequency candidate list, energy threshold screening and statistical significance test are performed, using the formula:

[0020]

[0021] Determine the key vibration frequency and generate a key vibration frequency list, where f represents the frequency point, E(f) represents the energy value of the frequency point f, S(f) represents the statistical significance index value of the frequency point f, ∈ is a constant, and γ is the threshold of the energy-significance ratio.

[0022] As a further solution of the present invention, the step of acquiring the vibration characteristic data set is:

[0023] Based on the key vibration frequency list, the formula is used:

[0024]

[0025] Calculate the amplitude value A(f) of each frequency point and generate a list of amplitude values, where P(f) represents the power spectrum density of frequency f and 2π is a normalization constant used to adjust the scale factor of amplitude calculation;

[0026] Performing pattern recognition on the amplitude value list to identify vibration patterns, including normal operating patterns and potential failure patterns, and outputting a classified set of vibration patterns;

[0027] The classified vibration mode set is integrated, key frequencies, amplitudes and classification labels of all vibration modes are recorded, and a vibration feature data set is constructed.

[0028] As a further solution of the present invention, the steps for constructing the mapping relationship between frequency and amplitude are:

[0029] The time series of each data point is extracted from the vibration feature dataset using the formula:

[0030]

[0031] Calculate the frequency value F(f) of each data point to generate spectrum data, where x(n) is the signal value at the nth time point, N is the total number of points in the time series, and f is the frequency point;

[0032] Analyze the spectrum data, identify key frequency components, and record peak values ​​and corresponding amplitudes to obtain frequency peak and amplitude pairing data;

[0033] The frequency peak and amplitude pairing data are used to capture the relationship between frequency and amplitude, and generate a mapping relationship table between frequency and amplitude.

[0034] As a further solution of the present invention, the step of obtaining the frequency response graph is:

[0035] According to the frequency and amplitude mapping relationship table, new vibration data is monitored in real time, and for each newly monitored frequency data point, the mapping table is queried to extract the corresponding amplitude value to generate real-time pairing data;

[0036] Based on the real-time paired data, digital signal processing is performed to remove noise and outliers in the data to obtain processed frequency and amplitude data;

[0037] The processed frequency and amplitude data are dynamically displayed to reveal the changes in vibration characteristics in a graphical manner and draw a frequency response diagram.

[0038] As a further solution of the present invention, the execution steps of the data comparison and historical trend analysis are:

[0039] Collect vibration sensor data in real time, fuse it with the frequency response graph, and integrate the two data forms to obtain a comprehensive data set;

[0040] Performing time series analysis on the comprehensive data set, extracting historical trends of the data, and obtaining historical trend analysis results;

[0041] Based on the historical trend analysis results, pattern matching and data comparison are performed to identify abnormal fluctuations in the data and obtain data comparison and trend analysis results.

[0042] As a further solution of the present invention, the steps for obtaining the engineering structure health analysis results are:

[0043] Based on the data comparison and trend analysis results, deviation analysis is performed to determine whether the deviation exceeds a preset threshold, abnormal points in the data are determined, and a preliminary deviation identification result is generated;

[0044] Performing cluster analysis on the preliminary deviation identification results, grouping the deviation data points according to vibration characteristics, identifying common abnormal patterns, including periodic faults and random faults, and generating cluster analysis results;

[0045] Using the cluster analysis results, all abnormal vibration modes and their potential impacts are recorded, maintenance recommendations are made, and engineering structure health analysis results are generated.

[0046] Compared with the prior art, the advantages and positive effects of the present invention are:

[0047] In the present invention, by continuously capturing video data of the surface of the engineering structure and extracting time series data from it, high-precision monitoring of subtle vibration fluctuations is achieved. Time-frequency analysis can more accurately identify key vibration frequencies and amplitudes. By constructing a mapping relationship between frequency and amplitude, subtle changes in vibration frequency can be effectively revealed, predictive maintenance and early fault identification can be performed, data reliability and analysis depth can be enhanced, and deviations and abnormal patterns can be comprehensively analyzed to identify them, thereby greatly improving the accuracy and timeliness of structural health monitoring and significantly reducing maintenance costs and downtime. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 is a system flow chart of the present invention;

[0049] Figure 2 This is a flow chart for obtaining vibration fluctuation signal data of the present invention;

[0050] Figure 3 This is a flow chart for identifying the key vibration frequency of the present invention;

[0051] Figure 4 A flowchart of obtaining a vibration characteristic data set of the present invention;

[0052] Figure 5 A flowchart for constructing a mapping relationship between frequency and amplitude of the present invention;

[0053] Figure 6 A flowchart for obtaining a frequency response diagram of the present invention;

[0054] Figure 7 The following is a flowchart of the data comparison and historical trend analysis of the present invention;

[0055] Figure 8 The figure is a flow chart for obtaining the health analysis results of the engineering structure of the present invention. DETAILED DESCRIPTION

[0056] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0057] In the description of the present invention, it should be understood that the terms "length", "width", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside" and the like indicate positions or positional relationships based on the positions or positional relationships shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, in the description of the present invention, "multiple" means two or more, unless otherwise clearly and specifically defined.

[0058] See also Figure 1 , the engineering vibration detection system based on machine vision includes:

[0059] The high-speed image capture module continuously records video data of the surface of the engineering structure through a camera, captures time series data from the video stream, and generates vibration fluctuation signal data;

[0060] The vibration feature extraction module performs time-frequency analysis on the signal based on the vibration fluctuation signal data, identifies the key vibration frequency, calculates the amplitude according to the key vibration frequency identification result, extracts and classifies the difference vibration mode, and obtains the vibration feature data set;

[0061] The frequency response mapping module calculates the frequency value of each data point based on the vibration characteristic data set, constructs the mapping relationship between frequency and amplitude, reveals the frequency change according to the mapping relationship, corresponds to the amplitude, and updates and draws the frequency response graph in real time;

[0062] The data fusion detection module fuses the frequency response graph with the vibration sensor data, performs data comparison and historical trend analysis, identifies deviations based on the analysis results, determines abnormal vibration modes, and generates engineering structure health analysis results.

[0063] Vibration fluctuation signal data includes vibration frequency data, time tags and spatial positioning information. Vibration feature data sets include frequency distribution records, amplitude analysis results and pattern recognition tags. Frequency response charts include frequency values, amplitude mapping results and real-time update charts. Engineering structure health analysis results include deviation identification results and abnormal vibration mode analysis records.

[0064] See also Figure 2 , the steps for obtaining vibration fluctuation signal data are:

[0065] Install cameras at key locations of engineering structures to capture surface changes in real time and generate real-time video stream data;

[0066] The process of configuring cameras at key locations of engineering structures involves determining the optimal camera installation angle and position in order to capture key dynamic changes in the structure. The technology for capturing surface changes in real time relies on high-resolution video capture devices that can record subtle structural changes. The steps to obtain a continuous video stream include using software for real-time transmission and storage of video streams, which requires sufficient storage space and good transmission bandwidth. In the process of generating real-time video stream data, the integrity and real-time nature of the data must be guaranteed to ensure the accuracy and effectiveness of subsequent analysis, and to obtain high-quality video data that can be directly used for analysis.

[0067] Perform image analysis on real-time video stream data, identify motion information and changes in each frame, extract time series, and obtain a time series data set;

[0068] The process of image analysis from real-time video streaming data begins with the preprocessing of video data, including noise reduction and resolution adjustment to improve the efficiency and accuracy of subsequent processing. Time series are extracted through edge detection and object tracking algorithms to identify motion information and changes in each frame. The algorithm needs to be adjusted to adapt to different environments and lighting conditions to ensure that key data can be accurately extracted from dynamic videos and a time series data set is obtained to ensure the reliability of the data and the accuracy of subsequent analysis.

[0069] Based on the time series data set, the formula is used:

[0070]

[0071] Calculate the change Δs between frames to obtain the vibration fluctuation signal data, where frame i and frame i-1 denote the i-th and i-1-th key frames respectively, and m denotes the total number of key frames;

[0072] There are three frames of data, each of which represents a digital quantification value of a structural change. The data are as follows:

[0073] The difference from frame 1 to frame 2 is quantized to 1000.

[0074] The difference quantized value from frame 2 to frame 3 is 1500.

[0075] The calculation process is as follows:

[0076]

[0077] The calculated frame-to-frame variation is 1802.78, which represents the magnitude of the vibration fluctuation signal extracted from the video data. This value reflects the magnitude of the dynamic changes of the structure caused by various external or internal factors during the observation period. Higher vibration fluctuation signal values ​​may indicate that the structure has undergone significant physical changes or has been subjected to greater external disturbances during the monitoring period, which is critical for engineering safety monitoring and maintenance, and can help engineers identify possible structural problems or areas that require further inspection.

[0078] See also Figure 3 , the steps to identify the key vibration frequency are:

[0079] Perform time-frequency conversion on the vibration fluctuation signal data, analyze the frequency components of the signal at the difference time points through short-time Fourier transform, and generate the signal time-frequency diagram;

[0080] Before performing spectrum analysis on the vibration fluctuation signal data, the data is first processed by a digital filter to filter out noise in the non-target frequency range and enhance the frequency band of interest in the signal. Butterworth or Chebyshev filters are used because of their different orders and types to adapt to different signal conditions. Subsequently, the data is subjected to a short-time Fourier transform (STFT). In this process, the signal is decomposed into a series of sine waves and cosine waves. Each wave corresponds to a specific frequency, and its amplitude and phase describe the contribution of the frequency in the original signal. The implementation of STFT is to obtain a spectrum by multiplying the complex exponential function point by point and accumulating it. The spectrum details the energy distribution from the lowest to the highest measurable frequency, providing a detailed view for further analysis, so that the subsequent frequency identification process can more accurately find those frequency components that dominate the structural response.

[0081] According to the signal time-frequency diagram, peak detection is performed to identify the frequency points where key energy is concentrated and generate a frequency candidate list;

[0082] After obtaining the spectrum distribution map, the next step is to use advanced signal processing technology to identify the key vibration frequency points in the map. The key to this step is to use the local peak detection algorithm, which locates the key frequency by searching for frequency points in the spectrum whose energy is significantly higher than the surrounding energy. To this end, a dynamic threshold is set, which is dynamically adjusted according to the average energy and standard deviation of the signal to adapt to the variability of different signals. The detection of critical frequencies not only depends on the energy size of the frequency, but also needs to consider its persistence in the spectrum and the energy comparison of adjacent frequency points to ensure that the identified frequency points are physically explainable and technically reliable. Each frequency point identified as critical will be recorded and used for further analysis, such as predicting possible structural defects or performance degradation through the amplitude and phase information of these frequency points, providing real-time monitoring and early warning of the structural health status.

[0083] Based on the frequency candidate list, energy threshold screening and statistical significance test are performed using the formula:

[0084]

[0085] Determine the key vibration frequency and generate a key vibration frequency list, where f represents the frequency point, E(f) represents the energy value of the frequency point f, S(f) represents the statistical significance index value of the frequency point f, ∈ is a constant to avoid zero division errors, and γ is the threshold of the energy-significance ratio;

[0086] In a certain actual measurement, for the frequency point f = 500Hz, the energy obtained is E(f) = 200 units of energy, the statistical significance S(f) = 20, and ∈ = 0.001 and γ = 0.05 are set. Then substitute into the formula for calculation:

[0087]

[0088] The results show that when the energy to statistical significance ratio at 500 Hz is greater than 0.05, the frequency point is identified as a critical vibration frequency. The judgment process helps to further screen out the main frequencies that affect the dynamic characteristics of the system, which is of great significance for system monitoring and maintenance.

[0089] See also Figure 4 , the steps to obtain the vibration feature data set are:

[0090] Based on the list of key vibration frequencies, the formula is used:

[0091]

[0092] Calculate the amplitude value A(f) of each frequency point and generate a list of amplitude values, where P(f) represents the power spectrum density of frequency f and 2π is a normalization constant used to adjust the scale factor of amplitude calculation;

[0093] At the frequency of 500 Hz, the power spectrum density is 50 units. Insert the specific values ​​for calculation:

[0094]

[0095] This shows that the amplitude at 500 Hz is about 2.82 units, which helps to assess the intensity of vibration at this frequency.

[0096] Perform pattern recognition on the amplitude value list to identify vibration modes, including normal operating modes and potential failure modes, and output a classified set of vibration modes;

[0097] After the amplitude data is obtained from the key frequency points, these amplitudes are classified through pattern recognition methods to identify different vibration modes. This process involves using machine learning algorithms, especially clustering algorithms such as K-means or hierarchical clustering, to analyze the amplitude data. Clustering algorithms classify data with similar amplitude values ​​into the same category by evaluating the similarity between each data point. Such classification helps identify common operating vibration modes and potential failure modes. For example, equipment that operates normally usually exhibits consistent vibration patterns, while equipment that fails may exhibit abnormal vibration patterns. This step not only includes the application of clustering algorithms, but also requires tuning of algorithm parameters, such as selecting the appropriate number of clusters and distance metrics to ensure the effectiveness and accuracy of clustering results. Through this analysis, valuable diagnostic information can be extracted from vibration data.

[0098] Integrate the classified vibration mode collection, record the key frequencies, amplitudes and classification labels of all vibration modes, and construct a vibration feature dataset;

[0099] Constructing a vibration feature dataset is a key step in integrating the analysis results. This dataset includes the classification information obtained from the previous steps. Each category contains different vibration frequencies, corresponding amplitudes, and identified vibration modes. The process of constructing this dataset involves the organization and formatting of data to ensure that each piece of data accurately reflects the actual situation of vibration. Each record in the dataset is based on actual measured vibration data, and the data is processed by advanced signal processing technology to ensure the quality and reliability of the data. In addition, the dataset will also be used for the training of machine learning models to predict and identify various vibration modes that may occur in the future. It not only helps to monitor the status of equipment in real time, but also takes preventive measures before potential problems occur, thereby reducing unplanned downtime and improving equipment efficiency and productivity.

[0100] See also Figure 5 , the steps to construct the mapping relationship between frequency and amplitude are:

[0101] The time series of each data point is extracted from the vibration feature dataset using the formula:

[0102]

[0103] Calculate the frequency value F(f) of each data point to generate spectrum data, where x(n) is the signal value at the nth time point, N is the total number of points in the time series, and f is the frequency point;

[0104] There is a simple time series data x(n)=[0, 1, 2, 3], and N=4.

[0105] Calculate F(1), the spectrum for f=1:

[0106] When n=0, x(0)=0 and the contribution is 0.

[0107] When n = 1, x(1) = 1, and the contribution is 1×e -i·2π·1 / 4 =1×(0.707-0.707i).

[0108] When n = 2, x(2) = 2, and the contribution is 2×e -i·2π·2 / 4 =2×(-1).

[0109] When n = 3, x(3) = 3, and the contribution is 3×e -i·2π·3 / 4 =3×(-0.707-0.707i).

[0110] F(1)=0+(0.707-0.707i)+(-2)+(-2.121-2.121i)

[0111] F(1)=-3.293-2.828i

[0112] The results show that the component with a frequency of 1 has a larger amplitude and a more complex phase, indicating that this frequency component contributes significantly to the original signal. This numerical result further reflects the complex changes of the original signal at frequency 1.

[0113] Analyze the spectrum data, identify the key frequency components, and record the peak values ​​and corresponding amplitudes to obtain the frequency peak and amplitude pairing data;

[0114] Analyze the frequency distribution in the spectrum data and identify the main frequency components. First, set an amplitude threshold and compare whether the amplitude of each frequency point exceeds this threshold to effectively screen out the significant frequency points in the signal. Then, record these main frequency points and their corresponding amplitude values ​​to provide basic data for establishing the mapping relationship. This process not only involves threshold comparison, but also includes classification of frequency points. Frequency points above the threshold are regarded as main frequency components. At the same time, the specific frequency value and relative amplitude of each main frequency point are recorded in detail to ensure the accuracy and reliability of these data, in preparation for constructing a detailed frequency and amplitude relationship diagram. These recorded data will be directly applied to the next step of mathematical modeling analysis.

[0115] The frequency peak and amplitude pairing data are used to capture the relationship between frequency and amplitude, and a mapping relationship table between frequency and amplitude is generated;

[0116] Using the collected frequency peak and amplitude pairing data, a mathematical model is constructed to describe the mapping relationship between frequency and amplitude. This modeling process uses a nonlinear regression method and calculates the parameters of the regression model through the least squares method based on the collected data points to ensure the prediction accuracy and generalization ability of the model. This model can predict the corresponding amplitude based on the input frequency. The construction of the model is not only based on statistical analysis, but also considers the rationality in the physical sense to ensure that the relationship between frequency and amplitude can accurately reflect the essence of actual physical phenomena. The final mapping relationship table provides a reliable basis for further analysis of vibration characteristics. The mapping relationship helps to better understand and predict the response behavior of the system at different frequencies.

[0117] See also Figure 6 , the steps to obtain the frequency response diagram are:

[0118] According to the frequency and amplitude mapping relationship table, new vibration data is monitored in real time. For each newly monitored frequency data point, the mapping table is queried to extract the corresponding amplitude value and generate real-time paired data;

[0119] The real-time monitoring system relies on the established mapping relationship table to respond to changes in frequency data. During this process, the system first detects new frequency data and immediately obtains the predicted amplitude value through high-speed query of the mapping relationship table. The query operation is based on an efficient data structure to optimize the retrieval speed to cope with frequent data updates and query needs. The mapping relationship table is obtained from historical data sets through machine learning model training, ensuring accurate mapping from frequency to amplitude. This real-time query and response process not only improves the system's ability to monitor dynamic changes, but also enables the relationship between frequency and amplitude to be updated in real time, providing real-time and accurate input data for subsequent data analysis and processing.

[0120] Based on the real-time paired data, digital signal processing is performed to remove noise and outliers in the data and obtain processed frequency and amplitude data;

[0121] In the process of processing the frequency and amplitude data acquired in real time, advanced digital signal processing technology is used. First, the sliding average algorithm is implemented to smooth the data to eliminate occasional noise and short-term fluctuations. In addition, dynamic thresholds are set to cut out outliers, which often come from sensor errors or external interference. The data processing at this stage not only improves the stability of the data, but also ensures the accuracy of data analysis, provides high-quality data for visualization, ensures the continuity and efficiency of the data processing process, and provides technical support for the system's abnormality detection and response.

[0122] For the processed frequency and amplitude data, the data is dynamically displayed, the changes of vibration characteristics are revealed graphically, and the frequency response diagram is drawn;

[0123] The processed data is transmitted to the graphical user interface (GUI) in real time for dynamic frequency response graph drawing. The visualization tool used can instantly display the relationship between frequency and amplitude. The real-time updated graphical display helps users intuitively understand and analyze the changes in vibration data. This chart is realized through advanced graphics drawing technology, including data-driven graphics rendering and dynamic scaling functions, which can handle large-scale data streams and display complex data relationships. This real-time graphical display is an indispensable part of the monitoring system. It not only improves the user interaction experience, but also enhances the real-time nature of data monitoring and the effectiveness of decision-making. The system can respond quickly to any vibration anomalies.

[0124] See also Figure 7 , the execution steps of data comparison and historical trend analysis are:

[0125] Collect vibration sensor data in real time, fuse it with the frequency response graph, and integrate the two data forms to obtain a comprehensive data set;

[0126] The real-time vibration data is fused with the existing frequency response diagram to form a comprehensive data set. The process involves integrating the sensor data with the frequency data, using data synchronization technology to ensure timestamp alignment, and unifying the format to ensure data type consistency. The comprehensive data set allows the analysis of time domain and frequency domain information under a unified framework. This fusion processing is to better reflect the current working status of the equipment. Through this method, the accuracy of subsequent analysis can be ensured, providing a data basis for identifying potential vibration anomalies.

[0127] Perform time series analysis on the comprehensive data set, extract historical trends of the data, and obtain historical trend analysis results;

[0128] Utilize comprehensive data sets to perform historical trend analysis, using time series analysis methods such as moving average and exponential smoothing techniques to extract and highlight the main trends in the data. The analysis process smoothes data fluctuations by calculating moving averages, and strengthens the weight of recent data through exponential smoothing methods to more sensitively capture possible changing trends. Technology helps separate real trends from random fluctuations, providing a basis for subsequent abnormal pattern recognition. By comparing the long-term and short-term behavior of data, future behavior patterns can be predicted, and identifying those data patterns that deviate from the norm is crucial for early diagnosis of potential equipment problems.

[0129] Based on the historical trend analysis results, perform pattern matching and data comparison to identify abnormal fluctuations in the data and obtain data comparison and trend analysis results;

[0130] Based on the data analysis and trend identification results of the previous steps, abnormal patterns are identified and engineering structure health analysis results are generated. This includes the use of statistical anomaly detection techniques such as Z-score and standard deviation analysis to identify abnormal fluctuations in the data, which can accurately mark abnormal points in the data, indicating that the equipment has potential failures. Then, combined with the historical maintenance records and engineering parameters of the equipment, the optimal maintenance plan is recommended, which not only improves the reliability of the equipment, but also optimizes the maintenance cost. Through comprehensive analysis results, engineers can adjust operating strategies in a timely manner and intervene in advance, thereby preventing serious equipment failures and ensuring production continuity and long-term operation of the equipment.

[0131] See also Figure 8 , the steps to obtain the results of engineering structure health analysis are:

[0132] Based on the data comparison and trend analysis results, deviation analysis is performed to determine whether the deviation exceeds the preset threshold, determine the abnormal points in the data, and generate preliminary deviation identification results;

[0133] Based on the results of data comparison and trend analysis, deviation analysis techniques are performed to identify anomalies in the data. This operation includes calculating the deviation between the real-time value of each data point and its historical average value. This historical average is calculated from the data collected in the past under the same operating conditions. Dynamic thresholds are set to evaluate these deviations. The thresholds are set based on past performance fluctuations and industry standards. Data points that exceed this threshold are considered potential anomalies. System deviation assessment helps accurately identify key data points that may indicate equipment performance degradation. Through precise deviation detection methods, preliminary deviation identification results can be generated, providing key input data for subsequent analysis steps.

[0134] Perform cluster analysis on the preliminary deviation identification results, group the deviation data points according to vibration characteristics, identify common abnormal patterns, including periodic faults and random faults, and generate cluster analysis results;

[0135] Perform in-depth cluster analysis on the data from the preliminary deviation identification results to clarify and classify the abnormal vibration patterns of the equipment. In this step, the data is grouped according to the deviation size and frequency characteristics of the data points, with the goal of identifying specific vibration abnormality patterns, such as periodic failures or random failures. By analyzing the similarities of the data points in each group, the possible failure types of the equipment can be identified. This analysis not only helps understand the current abnormal state of the equipment, but also predicts possible problems in the future. The generated cluster analysis results provide a scientific basis for equipment maintenance and fault prevention, enabling the maintenance team to develop more effective maintenance strategies for specific abnormal patterns.

[0136] Using cluster analysis results, record all abnormal vibration modes and their potential impacts, formulate maintenance recommendations, and generate engineering structural health analysis results;

[0137] The abnormal vibration modes identified by cluster analysis are integrated into the health analysis report of the engineering structure. This process includes a detailed analysis of the clustering results and converting them into specific maintenance recommendations. The report describes in detail the various abnormal vibration modes and their potential impact on the operation of the equipment. In addition, it also includes preventive measures and response strategies for each vibration mode. In this way, not only a comprehensive view of the current status of the equipment is provided, but also potential future problems are predicted. This allows the operation and maintenance team to adjust the operation strategy and maintenance plan in a timely manner based on the detailed information and recommendations in the report to extend the service life of the project.

[0138] The above are only preferred embodiments of the present invention and are not intended to limit the present invention in other forms. Any technician familiar with the profession may use the technical contents disclosed above to change or modify them into equivalent embodiments with equivalent changes and apply them to other fields. However, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present invention without departing from the technical solution of the present invention still falls within the protection scope of the technical solution of the present invention.

Claims

1. The engineering vibration detection system based on machine vision is characterized by: The system comprises: The high-speed image capture module continuously records video data of the surface of the engineering structure through a camera, captures time series data from the video stream, and generates vibration fluctuation signal data; The vibration feature extraction module performs time-frequency analysis on the signal based on the vibration fluctuation signal data, identifies the key vibration frequency, calculates the amplitude according to the key vibration frequency identification result, extracts and classifies the difference vibration mode, and obtains the vibration feature data set; The frequency response mapping module calculates the frequency value of each data point based on the vibration characteristic data set, constructs a mapping relationship between frequency and amplitude, reveals the frequency change according to the mapping relationship, corresponds to the amplitude, and updates and draws a frequency response graph in real time; The data fusion detection module fuses the frequency response diagram with the vibration sensor data, performs data comparison and historical trend analysis, identifies deviations based on the analysis results, determines abnormal vibration modes, and generates engineering structure health analysis results.

2. The machine vision-based engineering vibration detection system according to claim 1, characterized in that: The steps for obtaining the vibration fluctuation signal data are as follows: Install cameras at key locations of engineering structures to capture surface changes in real time and generate real-time video stream data; Performing image analysis on the real-time video stream data, identifying motion information and changes in each frame, extracting time series, and obtaining a time series data set; Based on the time series data set, the formula is adopted: Calculate the change Δs between frames to obtain the vibration fluctuation signal data, where frame i and frame i-1 They represent the i-th and i-1-th key frames respectively, and m represents the total number of key frames.

3. The machine vision-based engineering vibration detection system according to claim 2, characterized in that: The steps for identifying the key vibration frequency are: Performing time-frequency conversion on the vibration fluctuation signal data, analyzing the frequency components of the signal at the difference time points by short-time Fourier transform, and generating a signal time-frequency diagram; According to the signal time-frequency diagram, peak detection is performed to identify the frequency points where key energy is concentrated, and a frequency candidate list is generated; Based on the frequency candidate list, energy threshold screening and statistical significance test are performed using the formula: Determine the key vibration frequency and generate a key vibration frequency list, where f represents the frequency point, E(f) represents the energy value of the frequency point f, S(f) represents the statistical significance index value of the frequency point f, ∈ is a constant, and γ is the threshold of the energy-significance ratio.

4. The machine vision-based engineering vibration detection system according to claim 3, characterized in that: The steps of obtaining the vibration characteristic data set are: Based on the key vibration frequency list, the formula is used: Calculate the amplitude value A(f) of each frequency point and generate a list of amplitude values, where P(f) represents the power spectrum density of frequency f and 2π is the normalization constant; Performing pattern recognition on the amplitude value list to identify vibration patterns, including normal operating patterns and potential failure patterns, and outputting a classified set of vibration patterns; The classified vibration mode set is integrated, key frequencies, amplitudes and classification labels of all vibration modes are recorded, and a vibration feature data set is constructed.

5. The machine vision-based engineering vibration detection system according to claim 4, characterized in that: The steps for constructing the mapping relationship between frequency and amplitude are: The time series of each data point is extracted from the vibration feature dataset using the formula: Calculate the frequency value F(f) of each data point to generate spectrum data, where x(n) is the signal value at the nth time point, N is the total number of points in the time series, and f is the frequency point; Analyze the spectrum data, identify key frequency components, and record peak values ​​and corresponding amplitudes to obtain frequency peak and amplitude pairing data; The frequency peak and amplitude pairing data are used to capture the relationship between frequency and amplitude, and generate a mapping relationship table between frequency and amplitude.

6. The machine vision-based engineering vibration detection system according to claim 5, characterized in that: The steps of obtaining the frequency response diagram are: According to the frequency and amplitude mapping relationship table, new vibration data is monitored in real time, and for each newly monitored frequency data point, the mapping table is queried to extract the corresponding amplitude value to generate real-time pairing data; Based on the real-time paired data, digital signal processing is performed to remove noise and outliers in the data to obtain processed frequency and amplitude data; The processed frequency and amplitude data are dynamically displayed to reveal the changes in vibration characteristics in a graphical manner and draw a frequency response diagram.

7. The machine vision-based engineering vibration detection system according to claim 6, characterized in that: The execution steps of the data comparison and historical trend analysis are as follows: Collect vibration sensor data in real time, fuse it with the frequency response graph, and integrate the two data forms to obtain a comprehensive data set; Performing time series analysis on the comprehensive data set, extracting historical trends of the data, and obtaining historical trend analysis results; Based on the historical trend analysis results, pattern matching and data comparison are performed to identify abnormal fluctuations in the data and obtain data comparison and trend analysis results.

8. The machine vision-based engineering vibration detection system according to claim 7, characterized in that: The steps for obtaining the engineering structure health analysis results are as follows: Based on the data comparison and trend analysis results, deviation analysis is performed to determine whether the deviation exceeds a preset threshold, abnormal points in the data are determined, and a preliminary deviation identification result is generated; Performing cluster analysis on the preliminary deviation identification results, grouping the deviation data points according to vibration characteristics, identifying common abnormal patterns, including periodic faults and random faults, and generating cluster analysis results; Using the cluster analysis results, all abnormal vibration modes and their potential impacts are recorded, maintenance recommendations are made, and engineering structure health analysis results are generated.

Images