System and method for detecting cracks of structural component of carry-scraper in real time based on vibration characteristics

Abstract

The invention relates to the technical field of engineering mechanical structure health monitoring, and discloses a carry-scraper structural member crack real-time detection system and method based on vibration characteristics, and the carry-scraper structural member crack real-time detection system based on vibration characteristics comprises a sensing acquisition module which is used for acquiring structural member vibration response and obtaining an original signal; the signal processing module is used for carrying out wavelet denoising on the original signal and extracting effective vibration data; the feature extraction and dynamic calibration module is used for extracting multi-domain features and performing dynamic calibration in combination with working conditions; the crack recognition and intelligent diagnosis module is used for inputting an intelligent model to recognize a crack state; and the crack positioning and alarm feedback module is used for positioning cracks and giving an alarm in combination with the diagnosis result. According to the invention, the multi-modal sensor array is constructed, piezoelectric and MEMS sensors are cooperatively arranged, a magnetic snap-in type installation structure is combined, broadband response signals are stably collected under different working conditions, and then front-end signal optimization is completed in cooperation with a multi-scale wavelet denoising and standardization mechanism.

Description

Technical Field

[0001] The present invention relates to the technical field of engineering machinery structural health monitoring, and in particular to a real-time detection system and method for cracks in scraper structural parts based on vibration characteristics. Background Art

[0002] As underground mining operations progress toward deeper, higher-intensity operations, the structural safety of scrapers (Scrapers), core excavation equipment, has a direct impact on production continuity and the personal safety of operators. Scrapers are subjected to long-term alternating impact loads in confined spaces such as tunnels and intersections, making fatigue cracks highly susceptible to developing in structural components such as the frame, boom, and bucket. Once these cracks propagate, they are often difficult to detect in time, potentially leading to structural fracture, equipment downtime, and even accidents during heavy-duty operations. Therefore, there is an urgent need to develop a real-time monitoring system that can adapt to the high-interference environment underground and achieve early identification and precise location of microcracks in structural components.

[0003] Among existing structural damage detection technologies, static strain gauges and fiber Bragg grating sensors have been widely deployed on the surfaces of some key components to monitor changes in structural stress or displacement. They have the characteristics of fast response speed and high sensitivity to long-term chronic damage. At the same time, in terms of signal processing, existing technologies use first-order time domain indicators such as RMS and kurtosis as the main identification basis, and achieve state classification and simple alarms by setting fixed thresholds. At the application level, some systems have also introduced manual inspections and regular shutdown maintenance strategies, combined with visual inspections of the surface of structural parts, and have a certain ability to identify some obvious cracks. This type of method is simple to operate and the technology is mature. It has achieved certain application results in static or low-frequency operating equipment.

[0004] However, under complex dynamic working conditions, the above technical means have many bottlenecks. The visual inspection and downtime maintenance cycle is long and the response is delayed. Cracks can easily expand rapidly during operation, and the detection means lags seriously behind the damage evolution process. Although static strain gauges and fiber Bragg gratings can sense local responses, their limited frequency band makes it difficult to capture the high-frequency vibration modes induced by cracks. In addition, they have high installation requirements and poor environmental adaptability, and are prone to failure in underground vibration and dust scenarios. Most importantly, the recognition model that relies on a single eigenvalue and a fixed threshold has difficulty distinguishing between load fluctuations and damage-induced signal anomalies, which can easily lead to false alarms or missed alarms. There is a lack of dynamic feature correction mechanisms for different working conditions. The diagnostic reliability of existing systems under complex working conditions is generally low, making it difficult to meet the engineering requirements of online equipment status perception and fault closed-loop response. To this end, those skilled in the art have proposed a real-time detection system and method for cracks in scraper structural parts based on vibration characteristics to solve the above problems. Summary of the Invention

[0005] In response to the shortcomings of the existing technology, the present invention provides a real-time detection system and method for cracks in scraper structural parts based on vibration characteristics, which solves the problems of poor real-time crack identification, high false alarm rate, and inability to accurately locate in the existing technology.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solutions: a real-time detection system for cracks in scraper structural parts based on vibration characteristics, comprising:

[0007] The sensing acquisition module is used to collect the vibration response of the structural parts through sensors to obtain the original vibration signal;

[0008] a signal processing module, configured to perform wavelet threshold denoising on the original vibration signal to obtain a processed vibration signal;

[0009] a feature extraction and dynamic calibration module, configured to extract a multidimensional feature vector including time domain features, frequency domain features, and time-frequency domain features from the processed vibration signal, and calibrate the multidimensional feature vector according to real-time working conditions to obtain a calibrated feature vector;

[0010] a crack identification and intelligent diagnosis module, configured to input the calibrated feature vector into a preset intelligent diagnosis model for analysis to obtain a crack status diagnosis result of the structural component;

[0011] The crack location and alarm feedback module is used to determine the crack location according to the crack state diagnosis result and the original vibration signal, and generate an alarm signal.

[0012] Preferably, the sensing acquisition module includes:

[0013] Piezoelectric accelerometer, used to collect high-frequency vibration response signals;

[0014] MEMS inertial sensors are used to collect low-frequency structural response and attitude change information;

[0015] The mounting structure is used to mount the sensor on the metal surface of the vehicle frame, upper arm or connecting part by a combined fixing method of magnetic attraction and buckle.

[0016] Preferably, the signal processing module includes:

[0017] An adaptive sampling device for dynamically adjusting the sampling frequency based on the intrinsic frequency of the structural component;

[0018] A wavelet threshold denoising unit is used to perform multi-scale wavelet decomposition on the original vibration signal and perform an improved threshold function filtering operation on each scale coefficient;

[0019] The data normalization device is used to perform amplitude normalization processing on the vibration signal to eliminate the influence of equipment status differences.

[0020] Preferably, the multi-scale wavelet decomposition processing process in the wavelet threshold denoising unit includes:

[0021] Perform discrete wavelet decomposition on the original vibration signal to obtain multi-level decomposition coefficients;

[0022] At different scale layers, the corresponding adaptive thresholds are set according to the coefficient distribution characteristics;

[0023] Perform soft threshold processing on the decomposition coefficients of each layer to compress the noise component;

[0024] The processed coefficients are subjected to inverse wavelet transform to reconstruct the denoised vibration response signal.

[0025] Preferably, the feature extraction and dynamic calibration module includes:

[0026] A feature calculation unit, used to calculate multiple types of feature parameters of the vibration signal in the time domain, frequency domain and wavelet packet domain, wherein the feature parameters include root mean square, kurtosis, spectral centroid and energy entropy;

[0027] Working condition monitoring unit, used to collect real-time working status information of the scraper, including load, posture and motion status;

[0028] A dynamic correction unit is used to correct the frequency domain characteristics based on the operating condition information and adjust the characteristic threshold reference.

[0029] Preferably, the feature calibration process of the dynamic correction unit includes:

[0030] Obtain bucket load, equipment posture and working condition classification signals;

[0031] Establish a mapping relationship between the current working condition and the main characteristics in the frequency domain;

[0032] Correct the center frequency, energy amplitude and spectrum expansion index of the frequency domain feature according to the mapping relationship;

[0033] For the discriminant features with set thresholds, dynamic adjustments are made according to the degree of disturbance of the working conditions to generate a revised feature discrimination benchmark.

[0034] Preferably, the crack identification and intelligent diagnosis module includes:

[0035] a classification and recognition unit, configured to classify the calibrated feature vector using a trained convolutional neural network model and output a crack type of the structural component;

[0036] Multi-source fusion unit, used to fuse the diagnostic results of different sensors based on Bayesian reasoning method to generate consistency assessment results;

[0037] The discrimination output unit is used to output a status report including the diagnosis type, confidence level and feature deviation.

[0038] Preferably, the crack location and alarm feedback module includes:

[0039] Phase recognition unit, used to calculate the phase difference between multiple sensor nodes and infer the spatial position of the crack;

[0040] A grading and discrimination unit is used to classify crack risks based on the confidence level of the diagnosis results, the characteristic fluctuation amplitude, and the threshold offset;

[0041] The alarm execution unit is used to control the sound and light signal device to issue an alarm message and upload the alarm data to the vehicle terminal and remote monitoring platform.

[0042] The present invention also provides a real-time detection method for cracks in scraper structural parts based on vibration characteristics, comprising the following steps:

[0043] S1. Deploy the sensor acquisition module, place piezoelectric acceleration sensors and MEMS inertial sensors on the surface of the scraper structure, collect vibration response signals, and obtain original vibration signals;

[0044] S2. Performing wavelet threshold denoising and normalization processing on the original vibration signal to obtain a processed vibration signal;

[0045] S3. Extracting time domain, frequency domain and time-frequency domain features from the processed vibration signal to form a multi-dimensional feature vector;

[0046] S4. Collecting real-time working condition information of bucket load, equipment posture, and working condition classification signals, and dynamically correcting the multi-dimensional feature vector based on the information to obtain a calibrated feature vector;

[0047] S5. Inputting the calibrated feature vector into a convolutional neural network model to determine the crack state of the structural component;

[0048] S6. Calculate the crack spatial position based on multi-node phase difference and diagnostic confidence information;

[0049] S7. Output the crack status and positioning results, and trigger the sound and light alarm and remote information push.

[0050] Preferably, in step S6, the crack spatial position estimation process further includes:

[0051] Construct a spatial topological structure of sensor nodes and calculate the propagation speed based on the fixed distance between different nodes and the vibration propagation time difference;

[0052] The crack reflection source position is deduced by combining the phase difference and propagation velocity, and the position is fitted according to the node confidence weight to generate high-precision crack location coordinates.

[0053] The present invention provides a system and method for real-time detection of cracks in scraper structural parts based on vibration characteristics.

[0054] It has the following beneficial effects:

[0055] 1. The present invention constructs a multimodal sensor array and adopts the coordinated layout of piezoelectric and MEMS sensors, combined with a magnetic snap-on mounting structure, to stably acquire broadband response signals under different working conditions, and then cooperates with multi-scale wavelet denoising and normalization mechanisms to complete front-end signal optimization. This technical path realizes the stable acquisition and accuracy improvement of structural response signals in a high-noise background. Compared with the acquisition solutions in the existing technology that rely on a single sensor, are complex to install, and have poor environmental adaptability, it significantly improves the problems of poor signal integrity and sampling drift.

[0056] 2. The present invention proposes a multi-layer diagnostic architecture based on deep feature fusion and Bayesian probabilistic reasoning, inputs the dynamically calibrated multi-dimensional time-frequency features into a convolutional neural network, and introduces a confidence fusion mechanism for the results of multiple sensing channels to improve crack identification accuracy and diagnostic consistency. This method can form an early judgment on the state of microcracks. Compared with traditional strain or acceleration identification models based on threshold judgment, it solves the problems of poor recognition sensitivity and high false alarm rate in the low damage stage, and is particularly suitable for scenarios with degradation of recognition accuracy under complex load conditions.

[0057] 3. The present invention introduces a crack location method based on phase difference, and integrates the characteristic offset amplitude and model threshold drift ratio to construct a risk level judgment system. Through graded alarms and remote feedback, it realizes real-time closed-loop response to the damage status of structural parts. Compared with the existing technology that relies on manual inspections or only performs status identification without response strategies, this model solves the shortcomings of low positioning accuracy, delayed response, and inability to guide intervention, and enhances the practicality and system integration of online equipment health management. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 Schematic diagram of the system architecture of the present invention;

[0059] Figure 2 This is a schematic diagram of the sensor acquisition module architecture of the present invention;

[0060] Figure 3 This is a schematic diagram of the signal processing module architecture of the present invention;

[0061] Figure 4 Schematic diagram of the feature extraction and dynamic calibration module architecture of the present invention;

[0062] Figure 5This is a schematic diagram of the crack identification and intelligent diagnosis module architecture of the present invention;

[0063] Figure 6 This is a schematic diagram of the crack location and alarm feedback module architecture of the present invention;

[0064] Figure 7 Schematic diagram of the method of the present invention. DETAILED DESCRIPTION

[0065] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the drawings in the present specification. 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 making creative efforts are within the scope of protection of the present invention.

[0066] Please see the attached Figure 1 -Attached Figure 6 The embodiment of the present invention provides a real-time detection system for cracks in scraper structural parts based on vibration characteristics, comprising:

[0067] The sensing acquisition module is used to collect the vibration response of the structural parts through sensors to obtain the original vibration signal;

[0068] Specifically, in the system of the present invention, the vibration response signal of the structural component is the basic information source for subsequent crack diagnosis and state identification. In order to achieve accurate perception of the key load-bearing parts of the scraper under dynamic working conditions, the system is equipped with a sensor acquisition module for continuous, high-fidelity real-time sampling of structural vibrations during operation. This module cooperates with the subsequent signal processing module to build a complete crack feature extraction path. Especially in the context of underground mining equipment with strong vibration characteristics and complex working environments, this module must not only have high sensitivity and wide-band response capabilities, but also ensure sampling stability and environmental adaptability.

[0069] In this embodiment, the sensing acquisition module mainly includes a piezoelectric acceleration sensor, a MEMS inertial sensor, and a matching mounting structure.

[0070] Generally, scraper structural components exhibit significant multi-frequency vibration characteristics during operation. High-frequency components primarily originate from transient impacts or fatigue excitation, while low-frequency components are closely related to changes in overall structural stiffness and attitude fluctuations. Considering the importance of signals in different frequency bands, this embodiment employs a dual-type sensor deployment strategy.

[0071] Specifically, piezoelectric accelerometers are used to collect high-frequency vibration signals in the range of 0.5kHz to 10kHz, and are particularly suitable for monitoring the resonant response generated by local stiffness mutations caused by cracks. In some embodiments, the piezoelectric sensor is an ICP charge output sensor, whose output signal amplitude is proportional to the unit acceleration of the structure surface, and the sensitivity is generally 10-100mV / g, where g is the acceleration due to gravity, approximately equal to 9.81m / s. 2 .

[0072] Alternatively, a MEMS inertial sensor is used in this embodiment to supplement the acquisition of low-frequency response in the 1Hz to 200Hz range and triaxial attitude information. This sensor integrates a triaxial accelerometer and gyroscope module, enabling real-time capture of overall structural deformation trends caused by load changes.

[0073] In one possible implementation, the sensor sampling channel uses a differential input structure to suppress common-mode interference. The analog front end includes a bandpass filter whose passband frequency range is dynamically set based on the intrinsic frequency of the target structure. For example, if the first-order resonant frequency of the sampling arm is f1 = 52 Hz, the bandpass filter passband range can be set to [0.8f1, 2f1], or approximately [41.6, 104] Hz.

[0074] In order to ensure the adequacy of vibration signal sampling, the sampling frequency f s The Nyquist criterion must be met and further increased to more than ten times the upper frequency limit of the target signal, that is:

[0075] f s ≥max(10f c ,1 / Δt min );

[0076] Where: f c Indicates the structural eigenfrequency (Hz); Δt min is the minimum required time resolution (s).

[0077] This sampling strategy helps to avoid aliasing and ensure the multi-scale accuracy of subsequent wavelet processing.

[0078] To ensure stable signal sampling and coupling performance, this embodiment utilizes a combined magnetic and snap-on structure for sensor installation. The magnetic structure provides initial adhesion, ensuring adherence to the metal surface; the snap-on mechanism further defines its position and enhances shock resistance. This structure effectively prevents sensor dislodgement and poor signal contact in underground environments subject to frequent impact loads.

[0079] In some embodiments, to ensure accurate transmission of high-frequency response, the equivalent stiffness k of the sensor mounting structure is s The following conditions should be met:

[0080]

[0081] Where: k s is the installation structure stiffness (N / m); m s is the mass of the sensor (kg); ω max is the maximum angular frequency of the target monitoring frequency (rad / s), calculated as ω max =2πf max , where f max is the maximum frequency to be monitored (Hz).

[0082] This formula is used to ensure dynamic rigid coupling between the sensor and the structure to avoid high-frequency vibration transmission distortion.

[0083] Furthermore, in terms of sensor distribution strategy, this embodiment preferably places sensing nodes in high-stress concentration areas of the structure, such as the hinged connection between the frame and the boom, the bottom of the swing frame, and the rear wall of the bucket. These areas are more susceptible to microcrack initiation during stress cycles, and placing sensors helps to capture local modal changes in the structure as soon as possible.

[0084] As an expandable form, sensor nodes can synchronize and upload data through the CAN bus or RS-485 bus, providing a collaborative basis for system signal fusion and subsequent identification and diagnosis.

[0085] A signal processing module, used for performing wavelet threshold denoising on the original vibration signal to obtain a processed vibration signal;

[0086] Specifically, even after the vibration signal of a structural component is acquired by the sensing and acquisition module, problems such as background noise interference and amplitude drift still exist, making it difficult to directly use for crack identification and subsequent feature extraction. To improve signal quality and feature stability, the present invention provides a signal processing module, which serves as an intermediate link in the system and connects the two core processes of perception and analysis. This module primarily performs multi-scale wavelet denoising and amplitude normalization operations on the raw vibration signal to obtain clearer and standardized structural response data, providing stable input for the feature extraction module.

[0087] In this embodiment, the signal processing module includes an adaptive sampling device, a wavelet threshold denoising unit and a data normalization device.

[0088] In some embodiments, the adaptive sampling device sets sampling parameters based on the actual resonant frequency of the structural component to avoid information loss caused by low-frequency sampling while limiting data accumulation and computational burden caused by high-frequency data.

[0089] In general, the first-order eigenfrequency of the scraper structural components is f c= Fluctuates within the range of 30-80Hz. In order to adapt to the spectrum characteristics under variable working conditions, the sampling frequency f s Dynamic settings are:

[0090]

[0091] Where: f c Indicates the current resonant frequency of the structural component (unit: Hz), which can be obtained through spectrum peak identification or early calibration; Δt min The minimum time resolution required to represent the crack characteristic response (unit: s); f s is the actual sampling frequency (unit: Hz), which must be significantly larger than the upper limit of the target frequency to satisfy the Nyquist sampling theorem and provide higher wavelet decomposition capability.

[0092] As an option, the original vibration signal is first sent to the wavelet threshold denoising unit to perform multi-scale decomposition and noise suppression processing. In this embodiment, the discrete wavelet transform (DWT) is used to perform multi-layer decomposition of the signal. Specifically, the input signal x(t) is subjected to J-layer wavelet decomposition to obtain a set of detail coefficients d J (k) and approximation coefficient a J (k), where:

[0093]

[0094] The above coefficients reflect the vibration energy distribution in different frequency bands and can be used to identify spectrum anomalies caused by cracks.

[0095] In one possible implementation, for each layer coefficient w j,k The improved soft threshold function is used for noise reduction. The processing function is:

[0096]

[0097] Where: w j,k is the kth wavelet coefficient of the jth layer; j Indicates the initial threshold of the layer, which is determined by the coefficient standard deviation or median absolute deviation (MAD); σ j is the standard deviation of the coefficients of the jth layer.

[0098] Compared with the traditional soft threshold, this threshold function has better edge preservation ability and noise suppression performance, and is especially suitable for detecting weak crack signals.

[0099] After completing the threshold filtering for all scale coefficients, the inverse wavelet transform (IDWT) is performed to reconstruct the signal to obtain the denoised signal.

[0100] In some embodiments, in order to eliminate the amplitude reference differences caused by different sampling devices or installation positions, the signal is further sent to a data normalization device for normalization processing.

[0101] Specifically, the processed signal Normalized to zero mean and unit variance:

[0102]

[0103] Where: μ is the mean of the denoised signal (unit: original unit); σ is the standard deviation (unit: original unit).

[0104] This normalization form helps eliminate the impact of environmental fluctuations and device differences on subsequent feature analysis.

[0105] As a supplementary measure, if there is sensor channel drift or amplitude saturation anomaly, the present invention can introduce a sliding window dynamic normalization mechanism, using local statistics instead of the global mean variance to improve the system's robustness to sudden disturbances.

[0106] A feature extraction and dynamic calibration module is used to extract a multi-dimensional feature vector including time domain features, frequency domain features, and time-frequency domain features from the processed vibration signal, and calibrate the multi-dimensional feature vector according to the real-time working conditions to obtain a calibrated feature vector;

[0107] Specifically, after obtaining the denoised and normalized vibration response signal through the signal processing module, it is necessary to further extract physical characteristic parameters that can characterize the structural health state and perform dynamic corrections based on real-time operating conditions to enhance the accuracy and stability of the subsequent crack identification module. To this end, the present invention provides a feature extraction and dynamic calibration module for constructing a high-dimensional structural state representation vector and performing targeted calibration on it, ensuring good discrimination robustness under complex operating conditions.

[0108] In this embodiment, the feature extraction and dynamic calibration module includes a feature calculation unit, a working condition monitoring unit and a dynamic correction unit.

[0109] Generally speaking, the vibration response signals of structural components at different damage stages have identifiable statistical differences in the time domain, frequency domain, and time-frequency domain. Therefore, in order to improve the crack identification capability, the following three types of features are comprehensively extracted in this embodiment:

[0110] Specifically, in the time domain, indicators that reflect the amplitude characteristics and peak properties of the vibration waveform are selected.

[0111] include:

[0112] The root mean square value RMS is defined as:

[0113]

[0114] Where: x i Indicates the signal amplitude of the i-th sampling point (unit: m / s 2 ); N is the total number of points in each sampling signal (unit: dimensionless).

[0115] Kurtosis K is defined as:

[0116]

[0117] This indicator is used to measure whether the signal has abnormal spikes. Usually, crack initiation will cause the kurtosis to increase.

[0118] The form factor SF is defined as:

[0119]

[0120] in: is the absolute average value of the signal (unit: m / s 2 ).

[0121] As an option, frequency domain feature extraction uses the Fast Fourier Transform (FFT) method. After obtaining the spectrum X(f), the following features are extracted:

[0122] Spectral center frequency f sc , defined as:

[0123]

[0124] Where: f is the frequency variable (unit: Hz), |X(f)| 2 is the frequency domain amplitude square (power spectrum), f max is the upper limit of analysis frequency (unit: Hz).

[0125] Band energy ratio R band , defined as:

[0126]

[0127] Where: f1, f2 are the start and end boundaries of the target frequency band (unit: Hz). This indicator is used to identify changes in frequency band energy distribution.

[0128] In one possible implementation, the time-frequency domain features are obtained by wavelet packet energy analysis. The signal x(t) is decomposed by J-layer wavelet packets to obtain 2 J The energy value of the frequency band E i , based on which the energy entropy characteristics are constructed:

[0129]

[0130] Where: pi is the energy proportion of the i-th frequency band (unit: dimensionless); H represents the complexity of the signal frequency domain distribution (unit: nat).

[0131] In order to adapt to the influence of load fluctuation and posture disturbance on the characteristics under complex working conditions of the scraper, this embodiment further introduces a working condition monitoring unit and a dynamic correction unit to calibrate the feature vector.

[0132] In some embodiments, the working condition monitoring unit receives data streams from MEMS inertial sensors and load sensors. The monitored variables include:

[0133] Bucket load F (unit: kg),

[0134] Pitch angle θ (unit: °),

[0135] Peak vertical acceleration (Unit: m / s 2 ).

[0136] In the dynamic correction unit, the regression mapping function of frequency characteristics to working condition variables is established in combination with historical training data. For example, the correction expression of the spectral centroid frequency is:

[0137]

[0138] in: is the frequency of the center of gravity of the spectrum after calibration (unit: Hz); Δf(F) is the frequency shift correction caused by the load, defined as:

[0139] Δf(F)=α0+α1F+α2F 2 ;

[0140] Where: α0, α1, α2 are regression coefficients obtained by linear or polynomial regression fitting (unit: Hz, Hz / kg, Hz / kg 2 ).

[0141] In addition, for the classification threshold T0 of the set discriminant features (such as kurtosis, energy entropy), a dynamic adjustment strategy is used to generate a new threshold T adj :

[0142]

[0143] Where: β is the adjustment coefficient (unit: dimensionless); f ref is the reference frequency (unit: Hz), usually the first-order resonant frequency in the intact state.

[0144] In some embodiments, the dynamic correction process further stabilizes the update results through a sliding window averaging strategy to address the risk of miscalibration caused by short-term shocks or load disturbances.

[0145] The crack identification and intelligent diagnosis module is used to input the calibrated feature vector into the preset intelligent diagnosis model for analysis to obtain the crack status diagnosis results of the structural component;

[0146] Specifically, after completing feature extraction and dynamic calibration of the vibration signal, the present invention incorporates a crack identification and intelligent diagnosis module to further identify the presence and specific status of cracks in structural components. This module performs in-depth analysis of the calibrated multidimensional feature vectors and outputs a structural damage classification. A nonlinear mapping relationship is established between the feature layer and the decision layer, enhancing the system's ability to identify microcracks and complex damage states, and enabling comprehensive judgment based on multi-channel information.

[0147] In this embodiment, the crack identification and intelligent diagnosis module includes a classification and identification unit, a multi-source fusion unit, and a discrimination and output unit.

[0148] Generally, because the collected structural response signals exhibit strong nonlinear and non-stationary characteristics under different working conditions, traditional linear threshold discrimination models cannot accurately distinguish between normal and damaged states. To this end, this embodiment introduces a deep learning method to construct a classification and recognition model based on a convolutional neural network (CNN) to discriminate the dynamically calibrated feature vectors.

[0149] Specifically, the classification recognition unit receives the input feature vector v=[v1,v2,…,v n ], where v i represents the i-th calibrated feature parameter (the units correspond to the original units such as RMS, kurtosis, spectral center, etc.), and n is the feature dimension, usually 6–12 dimensions.

[0150] In one possible implementation, the CNN model includes an input layer, multiple convolutional layers, a pooling layer, a fully connected layer, and a Softmax output layer.

[0151] The convolution kernel size is set to 3×1 and the number of channels is 32-64 channels, which is used to extract the changes in the local pattern of the feature vector;

[0152] The pooling layer uses the maximum pooling operation (MaxPooling) for dimensionality reduction and compression;

[0153] The output layer uses the Softmax function to generate the classification probability vector P = [p0, p1, p2], where: p0 corresponds to the normal state probability; p1 corresponds to the microcrack state (crack depth <1mm) probability; p2 corresponds to the macrocrack state (crack depth ≥1mm) probability.

[0154] The Softmax function expression is as follows:

[0155]

[0156] Where: z i Represents the raw output score of the i-th category of the fully connected output layer (unit: dimensionless).

[0157] As an option, in order to enhance the stability and consistency of diagnosis, this embodiment provides a multi-source fusion unit for integrating classification results from multiple sensor nodes.

[0158] This unit performs confidence weighting processing based on the Bayesian inference method and outputs weighted fusion results.

[0159] In some embodiments, the diagnostic result from the kth sensing channel is

[0160] The corresponding prior credibility is set to w (k) ∈[0,1], satisfying:

[0161]

[0162] Where: K is the total number of channels.

[0163] The final fusion probability vector is defined as:

[0164]

[0165] The above fusion strategy can suppress channel information with low confidence and enhance the credibility of consistent diagnosis.

[0166] The discriminant output unit outputs the final status report based on the fusion result P = [P0, P1, P2], combined with the historical threshold distribution and feature offset information.

[0167] The report includes:

[0168] Crack state type: take the category corresponding to the maximum probability item;

[0169] Classification confidence level: defined as the maximum P i value;

[0170] Characteristic offset amplitude Δv j =v j -v j,ref , where: v j,ref It is the characteristic benchmark value under the reference state.

[0171] In some embodiments, if the maximum confidence max(P i)<θ0 (unit: dimensionless), the system can output an “uncertain” status prompt, indicating that manual review or delayed judgment is required.

[0172] In addition, to meet the continuous learning needs during equipment operation, the system supports an online incremental learning mechanism for the diagnostic model.

[0173] For samples (v, c) that have been manually confirmed, store them in the sample cache pool;

[0174] After every T sampling times, the model parameter fine-tuning process is triggered and the cached samples are used to perform local gradient updates;

[0175] The learning rate is set to η = 10 -4 ~10 -5 , to avoid rapid model drift.

[0176] This mechanism can adapt to the imperfect model in the early stage of equipment operation and gradually optimize the diagnostic accuracy.

[0177] The crack location and alarm feedback module is used to determine the crack location based on the crack status diagnosis results and the original vibration signal, and generate an alarm signal.

[0178] Specifically, after crack identification and state classification, the present invention incorporates a crack location and alarm feedback module to spatially locate cracks and promptly trigger an early warning mechanism. This module, the terminal functional unit of the system, primarily estimates crack location based on multi-point sensor response information, categorizes risk levels based on the confidence level of the diagnostic results and the magnitude of characteristic anomalies, and ultimately outputs an alarm signal that is synchronized to the human-machine interface and remote monitoring system. This output information is not only used for accident prevention and control but also provides a location reference for subsequent maintenance.

[0179] In this embodiment, the crack location and alarm feedback module includes a phase recognition unit, a classification determination unit and an alarm execution unit.

[0180] Generally, structural vibrations caused by cracks will experience phase disturbances in their spatial propagation characteristics. Especially when multiple sensor nodes are deployed, the same frequency component detected by different nodes will experience propagation time differences, allowing the approximate location of the vibration source on the structure to be inferred. This embodiment utilizes this characteristic to achieve preliminary crack location.

[0181] Specifically, the phase recognition unit constructs a crack location estimation model based on the phase difference between the signals collected at each node at the same frequency. Assume that the distance between nodes A and B is ΔL (unit: m), the main frequency of the vibration signal is f (unit: Hz), and the propagation phase velocity of the material in this frequency band is v p (unit: m / s), the phase difference between nodes is recorded as Δφ (unit: rad), then:

[0182]

[0183] The above formula can be used to reverse the phase difference Δφ and calculate the signal arrival time difference. Furthermore, by combining the phase relationships between multiple nodes, a spatial interpolation model can be constructed to achieve preliminary positioning of the crack hotspot.

[0184] As an option, to avoid positioning errors caused by boundary conditions or reflection interference, the phase recognition unit in this embodiment introduces a dynamic window function processing mechanism. That is, weighted filtering is performed on the original signal window, retaining only the stable frequency segment for positioning processing.

[0185] In some embodiments, in order to improve positioning accuracy, cross-fitting can be performed by combining phase difference data in multiple directions (such as vertical and longitudinal directions between the arm and the frame), and the final positioning point is based on the weighted result of the center of gravity of multiple fitting results.

[0186] After completing the positioning, the system enters the classification process. The classification unit performs risk classification based on the following parameters:

[0187] The feature offset amplitude Δv represents the offset of the key feature relative to the normal value;

[0188] Classification confidence P max , is the maximum category probability output by the recognition module;

[0189] The model threshold deviation ratio ρ is defined as:

[0190]

[0191] Where: T0 represents the basic threshold; T adj Indicates the real-time threshold after dynamic correction. The unit is consistent with the feature used.

[0192] In one possible implementation, the following multi-level discrimination strategy is set:

[0193] If Δv>δ2, P max >θ2, ρ>γ2, it is defined as level 3 alarm (high risk);

[0194] If only two of the above conditions are met, it is a Level 2 alarm (medium risk);

[0195] If only one of the conditions is met and the fluctuation is weak, a level 1 alarm (low risk) is triggered;

[0196] If none of the conditions are met, the situation is considered normal or enters the observation window period.

[0197] Among them: δ2, θ2, and γ2 are the setting limits of the offset threshold, confidence threshold, and parameter drift ratio, respectively. The specific values ​​can be determined by engineering debugging.

[0198] The alarm execution unit converts the above level output into actual control signal. As an option, the alarm information is triggered synchronously in the following two ways:

[0199] The controller outputs signals to drive the sound and light alarm device to alert the equipment personnel;

[0200] The crack location, level, timestamp and other information are packaged and synchronized to the vehicle terminal and remote platform via the CAN bus or wireless network.

[0201] In some embodiments, the alarm execution unit also converts the crack location information into coordinate points on a two-dimensional structural diagram of the equipment, and visually marks the abnormal location in a graphical manner, thereby improving the response efficiency of engineering personnel.

[0202] To avoid frequent false alarms, the module incorporates a sliding time window cumulative judgment mechanism during device operation. This mechanism requires three consecutive judgments of the same risk level within a set time window before a substantive alarm is triggered. This measure helps eliminate the impact of occasional noise or single diagnostic errors.

[0203] The real-time detection method for cracks in a scraper structural component based on vibration characteristics described below and the real-time detection system for cracks in a scraper structural component based on vibration characteristics described above can refer to each other.

[0204] Please see the attached Figure 7 The real-time detection method of cracks in scraper structural parts based on vibration characteristics includes the following steps:

[0205] S1. Deploy the sensor acquisition module, place piezoelectric acceleration sensors and MEMS inertial sensors on the surface of the scraper structure, collect vibration response signals, and obtain original vibration signals;

[0206] S2. Perform wavelet threshold denoising and normalization processing on the original vibration signal to obtain a processed vibration signal;

[0207] S3, extracting time domain, frequency domain and time-frequency domain features from the processed vibration signal to form a multi-dimensional feature vector;

[0208] S4, collecting real-time working condition information of bucket load, equipment posture and working condition classification signal, and dynamically correcting the multi-dimensional feature vector based on the information to obtain a calibrated feature vector;

[0209] S5. Inputting the calibrated feature vector into the convolutional neural network model to determine the crack state of the structural component;

[0210] S6. Calculate the crack spatial position based on multi-node phase difference and diagnostic confidence information;

[0211] S7. Output the crack status and positioning results, and trigger the sound and light alarm and remote information push.

[0212] The method of this embodiment can be used to execute the above system embodiment. Its principles and technical effects are similar and will not be described in detail here.

[0213] While 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 these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A real-time detection system for cracks in scraper structural parts based on vibration characteristics, characterized in that: include: The sensing acquisition module is used to collect the vibration response of the structural parts through sensors to obtain the original vibration signal; a signal processing module, configured to perform wavelet threshold denoising on the original vibration signal to obtain a processed vibration signal; a feature extraction and dynamic calibration module, configured to extract a multidimensional feature vector including time domain features, frequency domain features, and time-frequency domain features from the processed vibration signal, and calibrate the multidimensional feature vector according to real-time working conditions to obtain a calibrated feature vector; a crack identification and intelligent diagnosis module, configured to input the calibrated feature vector into a preset intelligent diagnosis model for analysis to obtain a crack status diagnosis result of the structural component; The crack location and alarm feedback module is used to determine the crack location according to the crack state diagnosis result and the original vibration signal, and generate an alarm signal.

2. The real-time detection system for cracks in scraper structural parts based on vibration characteristics according to claim 1 is characterized in that: The sensing acquisition module includes: Piezoelectric accelerometer, used to collect high-frequency vibration response signals; MEMS inertial sensors are used to collect low-frequency structural response and attitude change information; The mounting structure is used to mount the sensor on the metal surface of the vehicle frame, upper arm or connecting part by a combined fixing method of magnetic attraction and buckle.

3. The real-time detection system for cracks in scraper structural parts based on vibration characteristics according to claim 1 is characterized in that: The signal processing module includes: An adaptive sampling device for dynamically adjusting the sampling frequency based on the intrinsic frequency of the structural component; A wavelet threshold denoising unit is used to perform multi-scale wavelet decomposition on the original vibration signal and perform an improved threshold function filtering operation on each scale coefficient; The data normalization device is used to perform amplitude normalization processing on the vibration signal to eliminate the influence of equipment status differences.

4. The real-time detection system for cracks in scraper structural parts based on vibration characteristics according to claim 3 is characterized in that: The multi-scale wavelet decomposition processing process in the wavelet threshold denoising unit includes: Perform discrete wavelet decomposition on the original vibration signal to obtain multi-level decomposition coefficients; At different scale layers, the corresponding adaptive thresholds are set according to the coefficient distribution characteristics; Perform soft threshold processing on the decomposition coefficients of each layer to compress the noise component; The processed coefficients are subjected to inverse wavelet transform to reconstruct the denoised vibration response signal.

5. The real-time detection system for cracks in scraper structural parts based on vibration characteristics according to claim 1 is characterized in that: The feature extraction and dynamic calibration module includes: A feature calculation unit, used to calculate multiple types of feature parameters of the vibration signal in the time domain, frequency domain and wavelet packet domain, wherein the feature parameters include root mean square, kurtosis, spectral centroid and energy entropy; Working condition monitoring unit, used to collect real-time working status information of the scraper, including load, posture and motion status; A dynamic correction unit is used to correct the frequency domain characteristics based on the operating condition information and adjust the characteristic threshold reference.

6. The real-time detection system for cracks in scraper structural components based on vibration characteristics according to claim 5 is characterized in that: The characteristic calibration process of the dynamic correction unit includes: Obtain bucket load, equipment posture and working condition classification signals; Establish a mapping relationship between the current working condition and the main characteristics in the frequency domain; Correct the center frequency, energy amplitude and spectrum expansion index of the frequency domain feature according to the mapping relationship; For the discriminant features with set thresholds, dynamic adjustments are made according to the degree of disturbance of the working conditions to generate a revised feature discrimination benchmark.

7. The real-time detection system for cracks in scraper structural components based on vibration characteristics according to claim 1 is characterized in that: The crack identification and intelligent diagnosis module includes: a classification and recognition unit, configured to classify the calibrated feature vector using a trained convolutional neural network model and output a crack type of the structural component; Multi-source fusion unit, used to fuse the diagnostic results of different sensors based on Bayesian reasoning method to generate consistency assessment results; The discrimination output unit is used to output a status report including the diagnosis type, confidence level and feature deviation.

8. The real-time detection system for cracks in scraper structural components based on vibration characteristics according to claim 7 is characterized in that: The crack location and alarm feedback module includes: Phase recognition unit, used to calculate the phase difference between multiple sensor nodes and infer the spatial position of the crack; A grading and discrimination unit is used to classify crack risks based on the confidence level of the diagnosis results, the characteristic fluctuation amplitude, and the threshold offset; The alarm execution unit is used to control the sound and light signal device to issue an alarm message and upload the alarm data to the vehicle terminal and remote monitoring platform.

9. A method for real-time detection of cracks in a scraper structural component based on vibration characteristics, applied to a real-time detection system for cracks in a scraper structural component based on vibration characteristics according to any one of claims 1 to 8, characterized in that: The following steps are involved: S1. Deploy the sensor acquisition module, place piezoelectric acceleration sensors and MEMS inertial sensors on the surface of the scraper structure, collect vibration response signals, and obtain original vibration signals; S2. Performing wavelet threshold denoising and normalization processing on the original vibration signal to obtain a processed vibration signal; S3. Extracting time domain, frequency domain and time-frequency domain features from the processed vibration signal to form a multi-dimensional feature vector; S4. Collecting real-time working condition information of bucket load, equipment posture, and working condition classification signals, and dynamically correcting the multi-dimensional feature vector based on the information to obtain a calibrated feature vector; S5. Inputting the calibrated feature vector into a convolutional neural network model to determine the crack state of the structural component; S6. Calculate the crack spatial position based on multi-node phase difference and diagnostic confidence information; S7. Output the crack status and positioning results, and trigger the sound and light alarm and remote information push.

10. The real-time detection method for cracks in scraper structural components based on vibration characteristics according to claim 9, characterized in that: In step S6, the crack spatial position estimation process further includes: Construct a spatial topological structure of sensor nodes and calculate the propagation speed based on the fixed distance between different nodes and the vibration propagation time difference; The crack reflection source position is deduced by combining the phase difference and propagation velocity, and the position is fitted according to the node confidence weight to generate high-precision crack location coordinates.

Images