Construction machinery state monitoring method and system based on edge computing

Abstract

The application discloses an engineering machinery state monitoring method and system based on edge computing, relates to the technical field of state monitoring, and extracts an orthogonal characteristic sequence by performing orthogonal decomposition on an original vibration signal; under the condition that real-time rotating speed information is not accessed, extracts an instantaneous frequency of the orthogonal characteristic sequence, inversely reconstructs a virtual rotating speed sequence, constructs an adaptive amplitude envelope sequence based on a rotating speed ratio of a historical steady-state statistic of the orthogonal characteristic sequence and the virtual rotating speed sequence at a current moment, outputs stable normalized characteristic data, when the stable normalized characteristic data satisfies a preset deterioration triggering threshold, constructs Fourier substitute data according to the stable normalized characteristic data, takes a maximum spectrum peak on a preset fault order and a deviation degree of spectrum peak distribution of the Fourier substitute data spectrum peak as a significance test statistic, performs order component hypothesis testing to obtain a significance result, and distinguishes a real fault order from residual noise, and outputs a specific fault frequency band of an engineering machinery transmission component.

Description

Technical Field

[0001] This invention relates to the field of condition monitoring technology, and in particular to a method and system for condition monitoring of engineering machinery based on edge computing. Background Technology

[0002] In the field of construction machinery, the health status of transmission components directly determines the operational continuity and safety of large, heavy-duty equipment. With the development of the Industrial Internet of Things (IIoT) and edge computing technologies, the trend of pushing condition monitoring and fault diagnosis algorithms down to edge computing gateways closer to the sensor end to achieve low-latency real-time early warning of equipment anomalies has become an important industry trend. However, construction machinery is constantly subjected to harsh operating conditions of drastic speed changes, load changes, and strong physical collisions. On the one hand, transmission components often cannot be equipped with real-time speed sensors, causing traditional order tracking methods to fail; on the other hand, mechanical collisions generate high-energy transient impact noise, intertwined with deep frequency modulation and amplitude modulation phenomena, resulting in extremely complex non-stationary characteristics in the original vibration signal.

[0003] Currently, Chinese patent CN119437323A discloses a method, device, and engineering machinery for condition monitoring. By acquiring multi-dimensional sensor data collected by multiple sensors, the correlation between the data of multiple sensors is determined. The sensor data is iteratively trained and clustered using a neural network model of a self-organizing map network, and a condition monitoring feature piecewise linear graph reflecting the operating status of the engineering machinery is constructed. The existing technology has the following shortcomings: First, it relies heavily on multi-dimensional hardware sensors. In harsh environments, the failure of a single sensor can cause the entire condition monitoring model to collapse. Especially in the absence of high-precision real-time speed sensors, the existing technology completely loses the mechanical transmission operation benchmark, leading to direct diagnostic failure. Second, the pure data-driven algorithm is detached from the underlying physical constraints of the machinery and cannot physically separate normal transient physical impacts from the real steady-state cyclic degradation characteristics of transmission components. The existing algorithm is prone to misjudging sudden changes in working conditions caused by actions such as excavation collisions as mechanical failures, resulting in a very high false alarm rate. Third, construction machinery is accompanied by strong frequency and amplitude modulation under varying working conditions. The existing technology relies heavily on the computationally expensive equal-angle resampling algorithm to remove modulation interference, and the iterative process of the neural network model involved requires continuous consumption of huge computing and memory resources. Summary of the Invention

[0004] The technical problem solved by this invention is that existing technologies are unable to achieve effective signal purification and feature decoupling with low computing power consumption in harsh coupling environments where real-time speed sensors are missing and strong transient physical impacts and drastic changing working conditions are encountered. This results in traditional order analysis methods based on equal-angle resampling being unable to be deployed on edge gateways due to excessively high computational complexity. Furthermore, conventional pure data-driven diagnostic algorithms are highly susceptible to interference from both transient impact artifacts and speed amplitude modulation, resulting in extremely high false alarm rates. Consequently, they cannot achieve lightweight and highly robust real-time fault status localization of transmission components in engineering machinery.

[0005] To address the aforementioned technical problems, this invention provides the following technical solution: a method for monitoring the condition of engineering machinery based on edge computing, comprising the following steps:

[0006] Step S1: Collect the original vibration signal, construct a typical working condition redundancy dictionary for engineering machinery, perform orthogonal decomposition on the original vibration signal, and extract the orthogonal feature sequence.

[0007] Step S2: Without real-time speed information access, extract the instantaneous frequency of the orthogonal feature sequence, reconstruct the virtual speed sequence in reverse, construct an adaptive amplitude envelope sequence based on the ratio of the historical steady-state statistics of the orthogonal feature sequence to the speed of the virtual speed sequence at the current moment, and perform point-by-point amplitude division on the orthogonal feature sequence through the adaptive amplitude envelope sequence to output stable and standardized feature data.

[0008] Step S3: When the stationary standardized feature data meets the preset degradation trigger threshold, the stationary standardized feature data is used as input to construct Fourier substitution data. The deviation between the maximum spectral peak at the preset fault order and the distribution of the Fourier substitution data spectral peak is used as the significance test statistic to perform the order component hypothesis test.

[0009] Step S4: Based on the significance results of the order component hypothesis test, distinguish between the actual fault order and residual noise, and output the specific fault frequency band of the engineering machinery transmission component.

[0010] Preferably, step S1 includes the following sub-steps:

[0011] Step S101: Collect raw vibration signals using sensors deployed on the transmission components of the engineering machinery;

[0012] A typical operating condition redundancy dictionary for engineering machinery is constructed. The typical operating condition redundancy dictionary is composed of a steady-state operating condition sub-dictionary and a transient impact sub-dictionary. The steady-state operating condition sub-dictionary consists of M steady-state operating condition basis functions, and the transient impact sub-dictionary consists of N transient impact basis functions, where M and N are both positive integers.

[0013] Preferably, step S1 further includes the following sub-steps:

[0014] Step S102: Based on the typical working condition redundancy dictionary and the original vibration signal, the optimization objective is to minimize the norm of the joint sparse coefficients, and the constraint is that the reconstruction residual of the original vibration signal is less than a preset tolerance. A sparse objective optimization function is then constructed.

[0015] The sparse objective optimization function is iteratively approximated by the orthogonal matching pursuit algorithm to obtain the joint sparse coefficient vector corresponding to the original vibration signal, specifically including:

[0016] Based on the matrix block structure of the steady-state operating condition sub-dictionary and the transient impact sub-dictionary, the joint sparse coefficient vector is split into steady-state sparse coefficients corresponding to the steady-state operating condition sub-dictionary and transient sparse coefficients corresponding to the transient impact sub-dictionary.

[0017] Step S103: Multiply the steady-state condition sub-dictionary and the steady-state sparse coefficients to reconstruct a discrete vibration sequence containing only steady-state periodic components, and extract and output the discrete vibration sequence as an orthogonal feature sequence.

[0018] Discard the product terms of the transient shock sub-dictionary and transient sparse coefficients, and strip away the anomalous mutation data that matches the transient shock basis function.

[0019] Preferably, step S2 includes the following sub-steps:

[0020] Step S201: Under the condition of no real-time rotation speed information access, perform time-frequency transformation operation on the orthogonal feature sequence to obtain the time-frequency distribution matrix;

[0021] Extract the instantaneous frequency ridge of the preset fundamental order component from the time-frequency distribution matrix, and use the instantaneous frequency ridge as the instantaneous frequency of the preset fundamental order component.

[0022] Obtain the kinematic constraint proportional coefficient of the transmission component of the engineering machinery, and use the kinematic constraint proportional coefficient to perform a division operation on the instantaneous frequency of the preset basic order component to reconstruct a virtual speed sequence representing the current variable working condition.

[0023] Preferably, step S2 further includes the following sub-steps:

[0024] Step S202: Extract the effective values ​​of vibration amplitude of the orthogonal feature sequence within a preset historical steady-state time window, and use the effective values ​​of vibration amplitude as the statistics of the orthogonal feature sequence;

[0025] Calculate the ratio sequence of the virtual rotational speed sequence at the current moment and the average rotational speed within a preset historical steady-state time window;

[0026] Step S203: Multiply the rotational speed ratio sequence and the effective value of the vibration amplitude to construct an adaptive amplitude envelope sequence;

[0027] The adaptive amplitude envelope sequence is used as a normalization benchmark. The orthogonal feature sequence is then divided point by point using the normalization benchmark to output stationary and standardized feature data.

[0028] Preferably, step S3 includes the following sub-steps:

[0029] Step S301: When the stationary normalized feature data meets the preset degradation trigger threshold, the stationary normalized feature data is envelope demodulated, the envelope feature signal of the stationary normalized feature data is extracted, the envelope feature signal is Fourier transformed, the frequency axis is normalized to the order axis by the time average value of the virtual rotation speed sequence, and the original order amplitude spectrum and the original phase spectrum of the envelope feature signal are extracted.

[0030] Search for the maximum amplitude of the original order amplitude spectrum within the order range corresponding to the preset fault order, and take the maximum amplitude as the maximum spectral peak on the preset fault order.

[0031] Step S302: Generate multiple sets of independent random phase sequences that follow a uniform distribution, retain the original order amplitude spectrum unchanged, and replace the original phase spectrum with the random phase sequence;

[0032] The original order amplitude spectrum is retained and inverse Fourier transform is performed on multiple sets of random phase sequences to reconstruct multiple sets of time-domain sequences. The reconstructed sets of time-domain sequences are used as Fourier substitute data with retained power spectrum but randomized phase.

[0033] Preferably, step S3 further includes the following sub-steps:

[0034] Step S303: Under the premise that the transient impact and speed modulation have been eliminated from the stable and standardized feature data, there is no need to perform equal angle resampling operation. The Fourier transform operation is directly performed on the Fourier substitution data. The frequency axis is normalized to the order axis by using the time average value of the virtual speed sequence. The maximum substitution spectral peak of the Fourier substitution data in the order range corresponding to the preset fault order is extracted.

[0035] Calculate the statistical mean and statistical standard deviation of the largest spectral peak of the substitution, and construct the Fourier distribution of the spectral peaks of the substitution data;

[0036] Calculate the difference between the maximum spectral peak and the statistical mean at the preset fault order, divide the difference using the statistical standard deviation to obtain the deviation value, and use the deviation value as the deviation degree.

[0037] The degree of deviation was used as a significance test statistic for order component hypothesis testing.

[0038] Preferably, step S4 includes the following sub-steps:

[0039] Step S401: Perform a numerical comparison operation between the significance test statistic and the preset significance level threshold, and use the comparison result of the numerical comparison operation as the significance result of the order component hypothesis test.

[0040] When the significance result indicates that the significance test statistic is greater than the preset significance level threshold, the preset fault order is determined as the actual fault order.

[0041] When the significance result indicates that the significance test statistic is less than or equal to the preset significance level threshold, the preset fault order is determined as residual noise, thus completing the operation of distinguishing between the real fault order and residual noise.

[0042] Preferably, step S4 further includes the following sub-steps:

[0043] Step S402: When the preset fault order is determined to be the actual fault order, the bandwidth distribution range corresponding to the actual fault order is extracted, the bandwidth distribution range corresponding to the actual fault order is determined as the specific fault frequency band of the engineering machinery transmission component, a status warning data packet containing the specific fault frequency band is generated, and the status warning data packet is output.

[0044] The edge computing-based engineering machinery condition monitoring system includes a dictionary projection module, an adaptive demodulation module, an inspection module, and a positioning module.

[0045] The dictionary projection module is used to collect the original vibration signal, construct a redundant dictionary of typical working conditions of engineering machinery, perform orthogonal decomposition on the original vibration signal, and extract orthogonal feature sequences.

[0046] The adaptive demodulation module is used to extract the instantaneous frequency of the orthogonal feature sequence and reconstruct the virtual speed sequence in reverse when there is no real-time speed information access. Based on the ratio of the historical steady-state statistics of the orthogonal feature sequence to the speed of the virtual speed sequence at the current moment, an adaptive amplitude envelope sequence is constructed. The orthogonal feature sequence is then divided point by point using the adaptive amplitude envelope sequence to output stable and standardized feature data.

[0047] The testing module is used to construct Fourier substitution data by taking the stationary standardized feature data as input when the stationary standardized feature data meets the preset degradation trigger threshold. The deviation between the maximum spectral peak at the preset fault order and the distribution of the Fourier substitution data spectral peak is used as the significance test statistic to perform the order component hypothesis test.

[0048] The positioning module is used to distinguish between the actual fault order and residual noise based on the significance results of the hypothesis test of the order components, and output the specific fault frequency band of the transmission components of the engineering machinery.

[0049] The beneficial effects of this invention are as follows: By constructing a typical working condition redundant dictionary composed of a steady-state working condition sub-dictionary and a transient impact sub-dictionary, the original vibration signal is orthogonally decomposed using an orthogonal matching pursuit algorithm. This accurately removes abnormal and sudden data, adapts to edge hardware performance, and achieves extremely low computing power overhead. Virtual speed sequences are reconstructed even without real-time speed information access. An adaptive amplitude envelope sequence is constructed by multiplying the speed ratio sequence and the effective value of the vibration amplitude. Point-by-point amplitude division is performed using the adaptive amplitude envelope sequence. Without performing equal-angle resampling transformation, the speed amplitude modulation phenomenon caused by variable load conditions is completely eliminated, and stable standardized feature data is output. Fourier replacement data is constructed only when the stable standardized feature data meets a preset degradation trigger threshold, avoiding frequent and time-consuming replacement data checks during the equipment health phase. This greatly saves edge gateway computing and power resources. Finally, the significance test statistic is numerically compared with a preset significance level threshold to provide an objective quantitative decision boundary, effectively filtering out accidental spectral peak protrusions caused by complex working conditions, fundamentally reducing the false alarm rate in diagnosis. Attached Figure Description

[0050] Figure 1 A flowchart illustrating the steps of an edge computing-based engineering machinery condition monitoring method according to an embodiment of the present invention;

[0051] Figure 2 This is a basic flowchart of an edge computing-based engineering machinery condition monitoring system provided in one embodiment of the present invention. Detailed Implementation

[0052] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0053] Example 1, referring to Figure 1 This paper provides a method for monitoring the condition of engineering machinery based on edge computing, including the following steps:

[0054] Step S1: Collect the original vibration signal, construct a typical working condition redundancy dictionary for the engineering machinery. The typical working condition redundancy dictionary includes steady-state working condition basis functions and transient impact basis functions. Perform orthogonal decomposition on the original vibration signal and extract orthogonal feature sequences.

[0055] Step S2: Without real-time speed information access, extract the instantaneous frequency of the orthogonal feature sequence, reconstruct the virtual speed sequence in reverse, construct an adaptive amplitude envelope sequence based on the ratio of the historical steady-state statistics of the orthogonal feature sequence to the speed of the virtual speed sequence at the current moment, and perform point-by-point amplitude division on the orthogonal feature sequence through the adaptive amplitude envelope sequence to output stable and standardized feature data.

[0056] Step S3: When the stationary standardized feature data meets the preset degradation trigger threshold, the stationary standardized feature data is used as input to construct Fourier substitution data. The deviation between the maximum spectral peak at the preset fault order and the distribution of the Fourier substitution data spectral peak is used as the significance test statistic to perform the order component hypothesis test.

[0057] Step S4: Based on the significance results of the order component hypothesis test, distinguish between the actual fault order and residual noise, and output the specific fault frequency band of the engineering machinery transmission component.

[0058] This invention constructs a typical operating condition redundant dictionary by splicing together a steady-state operating condition sub-dictionary and a transient impact sub-dictionary. It employs an orthogonal matching pursuit algorithm to orthogonally decompose the original vibration signal, accurately removing anomalous and sudden data. This adapts to edge hardware performance and achieves extremely low computational overhead. Even without real-time speed information access, it reconstructs a virtual speed sequence. An adaptive amplitude envelope sequence is constructed by multiplying the speed ratio sequence and the effective value of the vibration amplitude. Point-by-point amplitude division is then performed using this adaptive amplitude envelope sequence. Without requiring equal-angle resampling transformation, the speed amplitude modulation phenomenon caused by variable load conditions is completely eliminated, outputting stable and standardized feature data. Fourier replacement data is constructed only when the stable and standardized feature data meets a preset degradation trigger threshold, avoiding frequent and time-consuming replacement data checks during equipment health phases. This significantly saves edge gateway computational and power resources. Finally, the significance test statistic is numerically compared with a preset significance level threshold, providing an objective quantitative decision boundary. This effectively filters out accidental spectral peaks caused by complex operating conditions, fundamentally reducing the false alarm rate in diagnosis.

[0059] In a specific embodiment, step S1 includes the following sub-steps:

[0060] Step S101: Collect raw vibration signals using sensors deployed on the transmission components of the engineering machinery.

[0061] A typical operating condition redundancy dictionary for engineering machinery is constructed. The typical operating condition redundancy dictionary is composed of a steady-state operating condition sub-dictionary and a transient impact sub-dictionary. The steady-state operating condition sub-dictionary consists of M steady-state operating condition basis functions, and the transient impact sub-dictionary consists of N transient impact basis functions, where M and N are both positive integers.

[0062] Step S102: Based on the typical working condition redundancy dictionary and the original vibration signal, the optimization objective is to minimize the norm of the joint sparse coefficients, and the constraint is that the reconstruction residual of the original vibration signal is less than the preset tolerance. A sparse objective optimization function is then constructed.

[0063] The sparse objective optimization function is iteratively approximated by the orthogonal matching pursuit algorithm to obtain the joint sparse coefficient vector corresponding to the original vibration signal, specifically including:

[0064] Based on the matrix block structure of the steady-state operating condition sub-dictionary and the transient impact sub-dictionary, the joint sparse coefficient vector is split into steady-state sparse coefficients corresponding to the steady-state operating condition sub-dictionary and transient sparse coefficients corresponding to the transient impact sub-dictionary.

[0065] Step S103: Multiply the steady-state condition sub-dictionary and the steady-state sparse coefficients to reconstruct a discrete vibration sequence containing only steady-state periodic components, and extract the discrete vibration sequence as an orthogonal feature sequence.

[0066] Discard the product terms of the transient shock sub-dictionary and transient sparse coefficients, and strip away the anomalous mutation data that matches the transient shock basis function.

[0067] In a preferred embodiment, for the orthogonal decomposition and dictionary reconstruction process, this invention employs a joint dictionary design based on sparse representation. The steady-state operating condition sub-dictionary is composed of discrete Fourier atoms based on different typical speeds of engineering machinery and the harmonic characteristics of different typical speeds, in order to efficiently extract the continuous periodic vibration of transmission components. The transient impact sub-dictionary is composed of Laplace wavelet atoms with different damping attenuation coefficients. The range of damping attenuation coefficients is limited to cover the low-damping range where the attenuation time is much longer than the fault impact period, which can match the unilateral exponentially decaying ringing response caused by non-periodic external transient physical impacts, while also retaining the periodic high-frequency impact modulation components caused by the fault.

[0068] The mathematical expression for the constructed sparse objective function is:

[0069] ;

[0070] in, The original vibration signal, This is a redundant dictionary for typical operating conditions of construction machinery. For the joint sparse coefficient vector, This is the preset tolerance.

[0071] Due to the basis Norm optimization is an NP-hard problem, and edge computing gateways use the orthogonal matching pursuit algorithm for iterative approximate solution. In each iteration, the orthogonal matching pursuit algorithm selects the dictionary atom most relevant to the current residual and projects the signal onto the orthogonal subspace spanned by the selected atom. This decomposes the original vibration signal into projected components on the orthogonal subspace spanned by the selected atom. By orthogonalizing the selected basis atoms in each iteration, the orthogonal matching pursuit algorithm solves the optimization problem with extremely low computational overhead, thus adapting to the hardware performance of edge devices.

[0072] In a specific embodiment, step S2 includes the following sub-steps:

[0073] Step S201: Without real-time rotational speed information, perform time-frequency transformation on the orthogonal feature sequence to obtain the time-frequency distribution matrix.

[0074] Extract the instantaneous frequency ridge of the preset fundamental order component from the time-frequency distribution matrix, and use the instantaneous frequency ridge as the instantaneous frequency of the preset fundamental order component.

[0075] Obtain the kinematic constraint proportional coefficients of the transmission components of the engineering machinery, and use the kinematic constraint proportional coefficients to perform a division operation on the instantaneous frequency of the preset basic order components to reconstruct a virtual speed sequence representing the current variable working condition.

[0076] Step S202: Extract the effective values ​​of vibration amplitude of the orthogonal feature sequence within a preset historical steady-state time window, and use the effective values ​​of vibration amplitude as the statistics of the orthogonal feature sequence.

[0077] Calculate the ratio of the virtual rotational speed sequence at the current moment to the average rotational speed within a preset historical steady-state time window.

[0078] Step S203: Multiply the rotational speed ratio sequence and the effective value of vibration amplitude to construct an adaptive amplitude envelope sequence that fluctuates in real time with changing operating conditions.

[0079] Using the adaptive amplitude envelope sequence as a normalization reference, the orthogonal feature sequence is subjected to point-by-point amplitude division using the normalization reference to complete amplitude standardization and output stationary standardized feature data free from speed amplitude modulation.

[0080] In a preferred embodiment, to acquire rotational speed information without the intervention of a rotational speed sensor hardware, the edge computing gateway first performs a time-frequency transformation operation on the orthogonal feature sequence. The time-frequency transformation operation preferably employs a short-time Fourier transform to obtain a time-frequency distribution matrix containing three-dimensional information, including time, frequency, and energy.

[0081] A ridge extraction algorithm is used to trace the frequency band trajectory with the most concentrated energy in the time-frequency distribution matrix along the time axis, and extract the instantaneous frequency of the most concentrated energy frequency band trajectory as a preset fundamental order component. If the monitoring target is a gearbox, the preset fundamental order component is preferably the gear meshing frequency; if the monitoring target is a bearing, the preset fundamental order component is preferably the bearing outer race fault characteristic frequency.

[0082] The kinematic constraint proportional coefficients of the transmission components of the engineering machinery are retrieved from pre-stored data. For gear transmission components, the kinematic constraint proportional coefficients are the known number of teeth on the gear. For rolling bearings, the kinematic constraint proportional coefficients are the characteristic frequency coefficients calculated based on geometric parameters, including the bearing pitch diameter, rolling element diameter, and contact angle. By dividing the extracted instantaneous frequency by the kinematic constraint proportional coefficient, the virtual speed sequence representing the current variable operating condition of the spindle can be accurately derived in reverse.

[0083] It should be noted that the edge computing gateway needs to extract the effective values ​​of the vibration amplitude of the orthogonal feature sequence within a preset historical steady-state time window, as the physical anchor point for subsequent amplitude modulation.

[0084] The preset historical steady-state time window is set based on the prior historical data collected during the startup and idling phase of the construction machinery. The preferred value of the preset historical steady-state time window is a continuous time segment with a duration of 2 to 5 seconds. The role of the preset historical steady-state time window here is to filter out transient fluctuation interference and provide a clean, stable, and physically meaningful reference energy system for subsequent dynamic amplitude scaling.

[0085] The edge computing gateway calculates the ratio of the virtual speed sequence at the current moment to the average speed within a preset historical steady-state time window. The speed ratio sequence reflects the deviation of the instantaneous speed from the steady-state reference speed under the current drastic operating conditions.

[0086] Since the vibration energy of transmission components in engineering machinery is usually positively correlated with the rotational speed, the edge computing gateway directly multiplies the rotational speed ratio sequence with the effective value of the vibration amplitude of the steady-state reference, thereby directly constructing an adaptive amplitude envelope sequence that fluctuates in real time with changing operating conditions in the time domain.

[0087] Using the adaptive amplitude envelope sequence as a normalization benchmark, the edge computing gateway performs point-by-point amplitude division on the orthogonal feature sequence using the normalization benchmark, i.e., point division operation. Through point-to-point division operation, the high amplitude caused by high speed at the current moment is reduced proportionally, and the low amplitude caused by low speed is amplified proportionally. Thus, without performing any computationally intensive equal-angle resampling transformation, the speed amplitude modulation phenomenon caused by variable load conditions is completely eliminated, and the output amplitude is flattened and standardized feature data is output.

[0088] In a specific embodiment, step S3 includes the following sub-steps:

[0089] Step S301: When the stationary normalized feature data meets the preset degradation trigger threshold, the stationary normalized feature data is envelope demodulated, the envelope feature signal of the stationary normalized feature data is extracted, the envelope feature signal is Fourier transformed, the frequency axis is normalized to the order axis by the time average value of the virtual rotation speed sequence, and the original order amplitude spectrum and the original phase spectrum of the envelope feature signal are extracted.

[0090] Specifically, Fourier transform is performed on the envelope feature signal to extract the frequency domain baseband spectrum. By dividing the frequency abscissa points of the frequency domain baseband spectrum by the time average value of the virtual rotation speed sequence, a linear scaling mapping of the frequency axis to the order axis is achieved, thereby extracting the original order amplitude spectrum and the original phase spectrum of the envelope feature signal.

[0091] The maximum amplitude of the original order amplitude spectrum is searched within the order range corresponding to the preset fault order, and the maximum amplitude is taken as the maximum spectral peak on the preset fault order.

[0092] Step S302: Generate multiple independent random phase sequences that follow a uniform distribution, retain the original order amplitude spectrum unchanged, maintain the original power spectrum characteristics, and replace the original phase spectrum with the random phase sequence.

[0093] The original order amplitude spectrum is retained and inverse Fourier transform is performed on multiple sets of random phase sequences to reconstruct multiple sets of time-domain sequences. The reconstructed sets of time-domain sequences are used as Fourier substitute data with retained power spectrum but randomized phase.

[0094] Step S303: Under the premise that the transient impact and speed modulation have been eliminated from the stable and standardized feature data, there is no need to perform equal-angle resampling operation. The Fourier transform operation is directly performed on the Fourier substitution data. The frequency axis is normalized to the order axis by using the time average value of the virtual speed sequence. The maximum substitution spectral peak of the Fourier substitution data in the order range corresponding to the preset fault order is extracted.

[0095] Calculate the statistical mean and statistical standard deviation of the largest spectral peak in the substitution, and construct the Fourier distribution of the spectral peaks in the substitution data.

[0096] Calculate the difference between the maximum spectral peak and the statistical mean at the preset fault level, divide the difference using the statistical standard deviation to obtain the deviation value, and use the deviation value as the degree of deviation.

[0097] The degree of deviation is used as the significance test statistic for order component hypothesis testing.

[0098] In a specific embodiment, step S4 includes the following sub-steps:

[0099] Step S401: Perform a numerical comparison operation between the significance test statistic and the preset significance level threshold, and use the comparison result of the numerical comparison operation as the significance result of the order component hypothesis test.

[0100] When the significance test statistic is greater than the preset significance level threshold, the preset fault order is determined as the actual fault order.

[0101] When the significance test statistic is less than or equal to the preset significance level threshold, the preset fault order is determined as residual noise, thus completing the operation of distinguishing between the true fault order and residual noise.

[0102] Step S402: When the preset fault level is determined to be the actual fault level, the bandwidth distribution range corresponding to the actual fault level is extracted, the bandwidth distribution range corresponding to the actual fault level is determined as the specific fault frequency band of the engineering machinery transmission component, a status warning data packet containing the specific fault frequency band is generated, and the status warning data packet is output.

[0103] In a preferred embodiment, the edge computing gateway does not constantly execute complex diagnostic algorithms. Instead, it uses a preset degradation trigger threshold for pre-control. This threshold is set based on the historical vibration envelope energy kurtosis or RMS trend curve of the transmission components of engineering machinery during full-lifecycle fatigue testing. The preferred value for this threshold is 1.5 to 2.0 times the historical statistical average under normal health baseline conditions. The preset degradation trigger threshold acts as a hardware trigger switch for the edge computing gateway to switch from low-power inspection mode to high-computing-power deep diagnostic mode, avoiding frequent, time-consuming alternative data checks during the device health phase, thereby significantly saving the edge gateway's computing and power resources.

[0104] When the data exceeds the preset degradation trigger threshold, the edge computing gateway performs envelope demodulation on the stable and standardized feature data. The envelope demodulation operation is preferably implemented by using a bandpass filter combined with Hilbert transform. By extracting the modulus of the analytical signal, the low-frequency fault envelope feature signal modulated on the high-frequency resonant carrier is accurately extracted.

[0105] In step S302, to construct a parameter-free statistical hypothesis testing benchmark, the edge computing gateway needs to generate multiple sets of Fourier alternative data. Independent and uniformly distributed random phase sequences are generated by uniform sampling within the continuous interval [0, 2π) using a pseudo-random number generator built into the edge computing gateway. By preserving the amplitude spectrum of the original envelope signal (i.e., preserving the power characteristics and autocorrelation structure of the original envelope signal) and forcibly scrambling the phase spectrum of the original envelope signal, the alternative data reconstructed by inverse Fourier transform can perfectly simulate pure random background noise with the same frequency domain energy distribution but completely free of any periodic fault impacts. Multiple sets of alternative data (preferably 100 sets) together constitute the statistical background distribution for subsequent testing.

[0106] In steps S303 and S4, after the edge computing gateway calculates the degree of deviation at the preset fault level (i.e., the statistical Z-score standard score, which is equal to the original spectral peak minus the mean of the surrogate data spectral peak, and then divided by the standard deviation), it needs to compare and judge the degree of deviation with the preset significance level threshold.

[0107] The preset significance level threshold is set based on the normal distribution confidence interval criterion in statistics (such as the Raida 3σ criterion) and the lower limit of the false alarm probability that construction machinery can tolerate in the field. The preferred value of the preset significance level threshold is 3 times the standard deviation (i.e., the Z-score threshold is 3). The role of the preset significance level threshold here is to provide an objective quantitative decision boundary with strict mathematical probability support, which can effectively filter out the random peak bulges caused by complex working conditions in the field, ensure that the final output of the true fault order has a high statistical confidence, and fundamentally reduce the false alarm rate of construction machinery under harsh working conditions.

[0108] When the deviation is greater than the preset significance level threshold (Z>3), it is determined that there is definite mechanical damage in the corresponding frequency band, such as broken gear teeth or peeling bearings. Then, a status warning data packet containing the specific fault frequency band is generated and uploaded to the cloud or local host computer. Otherwise, it is determined to be residual noise and no alarm is triggered.

[0109] Example 2, refer to Figure 2 It provides an edge computing-based engineering machinery condition monitoring system, including a dictionary projection module, an adaptive demodulation module, an inspection module, and a positioning module.

[0110] The dictionary projection module is used to acquire raw vibration signals, construct a redundant dictionary of typical working conditions for engineering machinery, perform orthogonal decomposition on the raw vibration signals, strip away abnormal mutation data matching transient impact basis functions, and extract orthogonal feature sequences.

[0111] The adaptive demodulation module is used to extract the instantaneous frequency of the orthogonal feature sequence and reconstruct the virtual speed sequence in reverse when there is no real-time speed information access. Based on the ratio of the historical steady-state statistics of the orthogonal feature sequence to the speed of the virtual speed sequence at the current moment, an adaptive amplitude envelope sequence is constructed. The orthogonal feature sequence is then divided point by point using the adaptive amplitude envelope sequence to output stationary and standardized feature data.

[0112] The testing module is used to construct Fourier substitute data by taking the stationary standardized feature data as input when the stationary standardized feature data meets the preset degradation trigger threshold. The deviation between the maximum spectral peak at the preset fault order and the distribution of the Fourier substitute data spectral peak is used as the significance test statistic to perform the order component hypothesis test.

[0113] The positioning module is used to distinguish between the actual fault order and residual noise based on the significance results of the order component hypothesis test, and to determine and output the specific fault frequency band of the transmission components of the engineering machinery.

[0114] This invention takes into account that, under conditions where there is no real-time speed information access, the original vibration signal of construction machinery is heavily interfered with by abnormal and sudden data, and is accompanied by strong speed amplitude modulation. At the same time, the computing power of edge computing devices is severely limited and prone to high false alarm rates. This invention innovatively abandons high-computing-power algorithms and external hardware dependence. First, a dictionary projection module is used to construct a redundant dictionary for typical working conditions to perform orthogonal decomposition on the original vibration signal, stripping away abnormal and sudden data that matches the transient impact basis function and extracting orthogonal feature sequences. To solve amplitude modulation interference, the instantaneous frequency of the orthogonal feature sequence is extracted by the adaptive demodulation module to reconstruct the virtual speed sequence in reverse. An adaptive amplitude envelope sequence is constructed based on the ratio of the historical steady-state statistics of the orthogonal feature sequence to the speed of the virtual speed sequence at the current moment. The orthogonal feature sequence is then subjected to point-by-point amplitude division in the pure time domain through the adaptive amplitude envelope sequence to output stable and standardized feature data. To avoid unnecessary computational waste, the testing module is activated only when the stationary standardized feature data meets a preset degradation trigger threshold. Fourier transform data is constructed using the stationary standardized feature data as input. The deviation between the maximum spectral peak at the preset fault order and the spectral peak distribution of the Fourier transform data is used as a significance test statistic to perform order component hypothesis testing. Finally, the localization module distinguishes between the actual fault order and residual noise based on the significance results, outputting the specific fault frequency band of the engineering machinery transmission component. This approach eliminates the constraints of external hardware speed sensors, transforming complex signal demodulation into a pure time-domain pointwise amplitude division process with extremely low computational cost. At resource-constrained edge environments, it completely eliminates speed amplitude modulation caused by varying operating conditions and thoroughly filters out accidental noise artifacts caused by complex operating conditions using rigorous statistical tests before the threshold is applied. This achieves lightweight and zero-false-report accurate localization of the fault frequency band of engineering machinery transmission components in harsh environments.

[0115] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the protection scope of the present invention.

Claims

1. A method for monitoring the condition of engineering machinery based on edge computing, characterized in that, Includes the following steps: Step S1: Collect the original vibration signal, construct a typical working condition redundancy dictionary for engineering machinery, perform orthogonal decomposition on the original vibration signal, and extract the orthogonal feature sequence. Step S2: Without real-time speed information access, extract the instantaneous frequency of the orthogonal feature sequence, reconstruct the virtual speed sequence in reverse, construct an adaptive amplitude envelope sequence based on the ratio of the historical steady-state statistics of the orthogonal feature sequence to the speed of the virtual speed sequence at the current moment, and perform point-by-point amplitude division on the orthogonal feature sequence through the adaptive amplitude envelope sequence to output stable and standardized feature data. Step S3: When the stationary standardized feature data meets the preset degradation trigger threshold, the stationary standardized feature data is used as input to construct Fourier substitution data. The deviation between the maximum spectral peak at the preset fault order and the distribution of the Fourier substitution data spectral peak is used as the significance test statistic to perform the order component hypothesis test. Step S3 includes the following sub-steps: Step S301: When the stationary normalized feature data meets the preset degradation trigger threshold, the stationary normalized feature data is envelope demodulated, the envelope feature signal of the stationary normalized feature data is extracted, the envelope feature signal is Fourier transformed, the frequency axis is normalized to the order axis by the time average value of the virtual rotation speed sequence, and the original order amplitude spectrum and the original phase spectrum of the envelope feature signal are extracted. Search for the maximum amplitude of the original order amplitude spectrum within the order range corresponding to the preset fault order, and take the maximum amplitude as the maximum spectral peak on the preset fault order. Step S302: Generate multiple sets of independent random phase sequences that follow a uniform distribution, retain the original order amplitude spectrum unchanged, and replace the original phase spectrum with the random phase sequence; The original order amplitude spectrum is retained and inverse Fourier transform is performed on multiple sets of random phase sequences to reconstruct multiple sets of time-domain sequences. The reconstructed sets of time-domain sequences are used as Fourier substitute data with power spectrum retained but phase randomized. Step S303: Under the premise that the transient impact and speed modulation have been eliminated from the stable and standardized feature data, there is no need to perform equal angle resampling operation. The Fourier transform operation is directly performed on the Fourier substitution data. The frequency axis is normalized to the order axis by using the time average value of the virtual speed sequence. The maximum substitution spectral peak of the Fourier substitution data in the order range corresponding to the preset fault order is extracted. Calculate the statistical mean and statistical standard deviation of the largest spectral peak of the substitution, and construct the Fourier distribution of the spectral peaks of the substitution data; Calculate the difference between the maximum spectral peak and the statistical mean at the preset fault order, divide the difference using the statistical standard deviation to obtain the deviation value, and use the deviation value as the deviation degree. The degree of deviation was used as a significance test statistic for order component hypothesis testing. Step S4: Based on the significance results of the order component hypothesis test, distinguish between the actual fault order and residual noise, and output the specific fault frequency band of the engineering machinery transmission component.

2. The engineering machinery condition monitoring method based on edge computing as described in claim 1, characterized in that, Step S1 includes the following sub-steps: Step S101: Collect raw vibration signals using sensors deployed on the transmission components of the engineering machinery; A typical operating condition redundancy dictionary for construction machinery is constructed. The typical operating condition redundancy dictionary is composed of a steady-state operating condition sub-dictionary and a transient impact sub-dictionary. The steady-state operating condition sub-dictionary is composed of M steady-state operating condition basis functions, and the transient impact sub-dictionary is composed of N transient impact basis functions, where M and N are both positive integers.

3. The edge computing-based engineering machinery condition monitoring method as described in claim 2, characterized in that, Step S1 further includes the following sub-steps: Step S102: Based on the typical working condition redundancy dictionary and the original vibration signal, the optimization objective is to minimize the norm of the joint sparse coefficients, and the constraint is that the reconstruction residual of the original vibration signal is less than a preset tolerance. A sparse objective optimization function is then constructed. The sparse objective optimization function is iteratively approximated by the orthogonal matching pursuit algorithm to obtain the joint sparse coefficient vector corresponding to the original vibration signal, specifically including: Based on the matrix block structure of the steady-state operating condition sub-dictionary and the transient impact sub-dictionary, the joint sparse coefficient vector is split into steady-state sparse coefficients corresponding to the steady-state operating condition sub-dictionary and transient sparse coefficients corresponding to the transient impact sub-dictionary. Step S103: Multiply the steady-state condition sub-dictionary and the steady-state sparse coefficients to reconstruct a discrete vibration sequence containing only steady-state periodic components, and extract and output the discrete vibration sequence as an orthogonal feature sequence. Discard the product terms of the transient impact sub-dictionary and transient sparse coefficients, and strip away the anomalous mutation data that matches the transient impact basis function.

4. The edge computing-based engineering machinery condition monitoring method as described in claim 3, characterized in that, Step S2 includes the following sub-steps: Step S201: Under the condition of no real-time rotation speed information access, perform time-frequency transformation operation on the orthogonal feature sequence to obtain the time-frequency distribution matrix; Extract the instantaneous frequency ridge of the preset fundamental order component from the time-frequency distribution matrix, and use the instantaneous frequency ridge as the instantaneous frequency of the preset fundamental order component. Obtain the kinematic constraint proportional coefficient of the transmission component of the engineering machinery, and use the kinematic constraint proportional coefficient to perform a division operation on the instantaneous frequency of the preset basic order component to reconstruct a virtual speed sequence representing the current variable working condition.

5. The edge computing-based engineering machinery condition monitoring method as described in claim 4, characterized in that, Step S2 further includes the following sub-steps: Step S202: Extract the effective values ​​of vibration amplitude of the orthogonal feature sequence within a preset historical steady-state time window, and use the effective values ​​of vibration amplitude as the statistics of the orthogonal feature sequence; Calculate the ratio sequence of the virtual rotational speed sequence at the current moment and the average rotational speed within a preset historical steady-state time window; Step S203: Multiply the rotational speed ratio sequence and the effective value of the vibration amplitude to construct an adaptive amplitude envelope sequence; The adaptive amplitude envelope sequence is used as a normalization benchmark. The orthogonal feature sequence is then subjected to point-by-point amplitude division using the normalization benchmark to output stationary and standardized feature data.

6. The engineering machinery condition monitoring method based on edge computing as described in claim 5, characterized in that, Step S4 includes the following sub-steps: Step S401: Perform a numerical comparison operation between the significance test statistic and the preset significance level threshold, and use the comparison result of the numerical comparison operation as the significance result of the order component hypothesis test. When the significance result indicates that the significance test statistic is greater than the preset significance level threshold, the preset fault order is determined as the actual fault order. When the significance result indicates that the significance test statistic is less than or equal to the preset significance level threshold, the preset fault order is determined as residual noise, thus completing the operation of distinguishing between the real fault order and residual noise.

7. The engineering machinery condition monitoring method based on edge computing as described in claim 6, characterized in that, Step S4 further includes the following sub-steps: Step S402: When the preset fault order is determined to be the actual fault order, the bandwidth distribution range corresponding to the actual fault order is extracted, the bandwidth distribution range corresponding to the actual fault order is determined as the specific fault frequency band of the engineering machinery transmission component, a status warning data packet containing the specific fault frequency band is generated, and the status warning data packet is output.

8. An edge computing-based engineering machinery condition monitoring system, applied in the edge computing-based engineering machinery condition monitoring method as described in any one of claims 1-7, characterized in that, It includes a dictionary projection module, an adaptive demodulation module, a verification module, and a positioning module; The dictionary projection module is used to collect the original vibration signal, construct a redundant dictionary of typical working conditions of engineering machinery, perform orthogonal decomposition on the original vibration signal, and extract orthogonal feature sequences. The adaptive demodulation module is used to extract the instantaneous frequency of the orthogonal feature sequence and reconstruct the virtual speed sequence in reverse when there is no real-time speed information access. Based on the ratio of the historical steady-state statistics of the orthogonal feature sequence to the speed of the virtual speed sequence at the current moment, an adaptive amplitude envelope sequence is constructed. The orthogonal feature sequence is then divided point by point using the adaptive amplitude envelope sequence to output stable and standardized feature data. The testing module is used to construct Fourier substitution data by taking the stationary standardized feature data as input when the stationary standardized feature data meets the preset degradation trigger threshold. The deviation between the maximum spectral peak at the preset fault order and the distribution of the Fourier substitution data spectral peak is used as the significance test statistic to perform the order component hypothesis test. The positioning module is used to distinguish between the actual fault order and residual noise based on the significance results of the hypothesis test of the order components, and output the specific fault frequency band of the transmission components of the engineering machinery.

Images