Crane slewing bearing work monitoring method and system and application

Abstract

The invention discloses a crane slewing bearing work monitoring method and system and application, according to the scheme, a sound wave data set is ingeniously used as the basis, after data augmentation processing is carried out on the sound wave data set, an augmented data set having certain relevance with sound wave data of the sound wave data set is obtained, and therefore during follow-up neural network training, the sound wave data of the sound wave data set is subjected to data augmentation processing; the detection neural network obtained by training has better detection reliability on the sound wave data of the slewing bearing; on the basis, according to the scheme, when the detection neural network is used for detecting the monitoring information of the crane, after signal denoising, time-frequency analysis and deep learning network processing are carried out on the collected acoustic wave signals, acoustic wave data with less noise and better characteristic signal trueness can be obtained; and better high-quality input data is provided for detecting reliability application of the neural network.

Description

Technical Field

[0001] The present invention relates to the fields of crane monitoring technology and deep learning network application technology, and in particular to a crane slewing bearing operation monitoring method, system and application. Background Art

[0002] As a popular heavy-duty transport equipment, cranes have the advantages of strong transport capacity and stable equipment use, so they are widely used in production workshops, storage workshops, ports, construction sites, etc. As an important component of cranes, the work safety and reliability research of slewing bearings has always been a hot topic in this field.

[0003] Among them, the slewing bearing is also called the turntable bearing. It has a wide range of applications in real industry, so it is often called "the joint of the machine". The slewing bearing is an important transmission component required for machinery that needs to make relative rotational motion between two objects and simultaneously bear axial force, radial force, and tipping moment. With the rapid development of the machinery industry, slewing bearings are widely used in cranes, ship equipment, engineering machinery, light industrial machinery, metallurgical machinery, medical machinery, industrial machinery and other industries.

[0004] Due to the complex working environment of the crane, the slewing bearing may be disturbed by a variety of external and internal factors during the process of being driven. Due to the complex and changeable working environment of the slewing bearing during service, there are many factors that affect its service life. Most of the existing simulation model-based or data-driven methods train the deep learning network through the training set and use it to detect whether the slewing bearing has faults or service life. In practical applications, the data of the training set is mostly idealized or the number base is not large enough, which makes the model often cause estimation errors due to excessive signal interference that causes slewing bearing failure and end of service life. At the same time, the current full life cycle monitoring data of the slewing bearing is difficult to reuse and learn from due to statistical factors and small data volume. Therefore, how to improve the diversity and reliability of the training data of the deep learning model and combine scientific and reasonable abnormal detection schemes and early warning schemes to ensure the safe and reliable operation of the crane is a very positive and realistic topic. Summary of the invention

[0005] In view of this, the purpose of the present invention is to propose a crane slewing bearing operation monitoring method, system and application with reliable and sufficient training samples, flexible application and reliable implementation.

[0006] In order to achieve the above technical objectives, the technical solution adopted by the present invention is:

[0007] A method for monitoring the operation of a slewing bearing of a crane, comprising:

[0008] S01, collecting the acoustic wave data of the slewing bearing of the crane under different working conditions and its corresponding working parameters, then annotating the acoustic wave data according to the working conditions, and then associating the acoustic wave data and its corresponding working parameters and information annotations to form an acoustic wave data group;

[0009] S02, obtaining an acoustic wave data set, amplifying it according to preset conditions to obtain an amplified data set, and then combining the acoustic wave data set and the amplified data set to form an acoustic wave data set;

[0010] S03, extracting data from the acoustic wave data set, dividing the data into training data and verification data, and then inputting the training data into the neural network for training, and then verifying the trained neural network through the verification data until the model converges, thereby obtaining a detection neural network for abnormality detection of the crane slewing bearing;

[0011] S04, responding to the work start signal of the crane, monitoring the working parameters of the slewing bearing of the crane in real time, and collecting the acoustic wave data of the slewing bearing when it is working in real time, and then compiling it to form monitoring information;

[0012] S05, inputting the monitoring information into the detection neural network, and the detection neural network outputs the detection result;

[0013] S06. Obtain the detection result. When the detection result indicates that the slewing bearing of the crane is in an abnormal condition, generate and output a warning message.

[0014] As a possible implementation, further, in the present solution S01, the source of the collected sound wave data is one or more of the following:

[0015] (1) Acoustic wave data collected from the crane's slewing bearing during actual operation;

[0016] (2) constructing a slewing bearing working platform based on the laboratory, making the slewing bearing reach a preset working state according to preset working parameters, and then collecting data to obtain acoustic wave data;

[0017] The information marking includes whether the slewing bearing working condition is abnormal and the type of abnormality;

[0018] The working parameters include: the slewing speed, the rotation amplitude and the working environment temperature of the slewing bearing.

[0019] As a preferred implementation option, preferably, in the present solution S02, the method for amplifying the acoustic wave data set includes time domain data augmentation processing, frequency domain data augmentation processing, hybrid data augmentation processing or augmentation processing through a deep learning network;

[0020] After the acoustic wave data set is amplified, the working parameters and information annotations of the amplified data set obtained are consistent with the working parameters and information annotations of the corresponding acoustic wave data set; the amplified data set also has an annotation marking it as amplified data.

[0021] As a preferred implementation option, preferably, the time domain data augmentation processing described in this solution includes time expansion processing or time shift processing of the sound wave data;

[0022] Time stretching is to amplify the data by changing the time scale of the sound wave data, that is, to obtain the amplified data by compressing or stretching the time axis of the signal without changing the frequency characteristics of the sound wave signal, which includes: using linear interpolation or phase vocoder to adjust the scale factor α of the time axis, and its formula is defined as follows:

[0023]

[0024] Among them, x new (t) represents the sound wave data after time expansion and contraction processing, x(t) represents the initial sound wave data, when α is greater than 1, it represents time expansion, and when α is less than 1, it represents time compression;

[0025] Time shift processing is to translate the sound wave data on the time axis, increase diversity while keeping the frequency characteristics of the signal unchanged, by translating the entire signal forward or backward by a certain time offset δt, which is defined as follows:

[0026] x new (t) = x(t+δt)

[0027] Among them, x new (t) represents the sound wave data after time shift processing, x(t) represents the initial sound wave data, δt represents the translation time, and a negative value represents a backward translation, and a positive value represents a forward translation.

[0028] As a preferred implementation option, preferably, the frequency domain data augmentation processing described in this solution includes adding frequency disturbance to the sound wave signal in the frequency domain and performing spectrum trimming on the sound wave signal;

[0029] Frequency perturbation is added to the sound wave signal in the frequency domain to ensure that the diversity of the data is enhanced while maintaining the main frequency characteristics. The frequency domain characteristics are perturbed by adding noise or fine-tuning the spectrum. The formula is defined as follows:

[0030] x new (ω)=X(ω)+ε(ω)

[0031] Among them, ε(ω) is the frequency domain disturbance noise, X(ω) is the sound wave data, and x new(ω) is the acoustic wave data after adding frequency disturbance;

[0032] The spectrum trimming of the sound wave signal is to retain the main part of the signal spectrum, remove or attenuate the secondary frequency components, so as to generate a variety of frequency domain samples; by windowing the signal spectrum, it only retains the signal within a certain frequency range. The formula is defined as follows:

[0033] x new (ω)=X(ω)·W(ω)

[0034] Among them, W(ω) is the window function, X(ω) is the sound wave data, x new (ω) is the sound wave data after spectrum trimming.

[0035] As a preferred implementation option, preferably, the mixed data augmentation processing in this solution includes additive noise processing or mixed sample processing;

[0036] The method for augmenting processing by a deep learning network includes augmenting samples by a generative adversarial network or augmenting samples by an autoencoder and a variational autoencoder;

[0037] Additive noise processing is to increase the diversity of the data set by adding a small amount of noise to the original signal. It includes adding noise n(t) in the time domain or frequency domain of the sound wave data. The formula is defined as follows:

[0038] x new (t) = x(t) + n(t)

[0039] Among them, x new (t) is the sound wave data after adding noise, x(t) is the sound wave data, and n(t) is Gaussian noise or uniform noise;

[0040] The mixed sample processing is to mix two different samples to generate a new sample to enhance the robustness of the model to different signal modes, which is defined as follows:

[0041] x new (t) = λx1(t) + (1-λ)x2(t)

[0042] Among them, x new (t) is the mixed sound wave data, x1(t) and x2(t) are different sound wave data, and λ is a random number between [0, 1], which is used to balance the mixing ratio of the two signals;

[0043] Augmenting samples through generative adversarial networks is to build GANs containing generators and discriminators, and then generate adversarial training between them and the discriminator to generate augmented samples that meet the requirements;

[0044] Augmenting samples through autoencoders and variational autoencoders is the encoding and decoding of original data to generate augmented samples with similar characteristics to the original samples. It includes training the autoencoder network, then compressing the data into the latent space, and reconstructing the decoder to generate new samples. The formula is defined as follows:

[0045] z=f encoder (x)

[0046] x new =f encoder (z+∈)

[0047] Among them, z is the implicit space representation, ∈ is the added random noise, and x new is the new sample generated.

[0048] As a preferred implementation option, preferably, in the present solution S04, when collecting the acoustic wave data of the slewing bearing during operation, the collected signals are also subjected to signal denoising, time-frequency analysis and deep learning network processing to remove interference in the signals, which includes:

[0049] (1) Signal denoising

[0050] Use a bandpass filter to denoise the collected acoustic wave signal to remove low-frequency and high-frequency interference signals so that the signal is concentrated in the frequency band containing fault information; assuming that the original signal is x(t) and the filtered signal is y(t), it can be expressed as:

[0051] y(t)=x(t)*h(t)

[0052] Where h(t) is the impulse response function of the filter, and * represents the convolution function;

[0053] (2) Time-frequency analysis

[0054] The short-time Fourier transform (STFT) is used to transform the signal from the time domain to the time-frequency domain to capture the transient characteristics of the signal. The formula is defined as follows:

[0055]

[0056] Among them, X(t,ω) is the complex representation in the time-frequency domain, which is the STFT result, that is, the complex value representation of the signal at time t and frequency ω, which represents the amplitude and phase information of the signal at ω within a small time window at time t; x(τ) is the time domain representation of the original signal x(t), τ represents the time variable, ω(τ-t) is the time window function, which weights the signal within a period of time near time t, τ in the time window function represents the time variable, ω is the frequency variable, e -jωτis a complex exponential function, which is the core part of Fourier transform and is used to convert signals from time domain to frequency domain. j in the complex exponential function is an imaginary unit, which represents the imaginary part of a complex number. dτ is an integral variable, which represents the integration of time τ. The integration is to accumulate the frequency components of the signal weighted by the window function on the entire time axis.

[0057] (3) Convolutional Neural Network (CNN) processing

[0058] The spatial features of time-frequency data are extracted through the convolutional neural network CNN. The time-frequency feature matrix X(t,ω) obtained by time-frequency analysis is used as input. The output of the lth convolutional layer of the convolutional neural network CNN can be expressed as:

[0059]

[0060] in, is the output feature at position (i, j); is the weight matrix of the convolution kernel, the size of the convolution kernel is k×k; b l is the bias term, σ(·) is the activation function;

[0061] (4) Long Short-Term Memory Network (LSTM) Processing

[0062] The convolution kernel of the convolutional neural network CNN is used to extract features layer by layer, and the extracted features are used as input items of the long short-term memory network LSTM; the long short-term memory network LSTM is used to capture the time dependency of the data, which includes the following calculation steps:

[0063] Forget gate calculation:

[0064] f t =σ(W f ·[h t-1 , x t ]+b f )

[0065] Input gate calculation:

[0066] i t =σ(W i ·[h t-1 , x t ]+b i )

[0067] Candidate memory cell state calculation:

[0068]

[0069] Memory unit status update:

[0070]

[0071] Output gate calculation and final output:

[0072] o t =σ(W o ·[h t-1 , x t ]+b o )

[0073] h t =o t ⊙tanh(C t )

[0074] Among them, x t is the input feature sequence of the long short-term memory network LSTM; h t is the output state of the long short-term memory network LSTM; W f , W i , W C , W o is the weight matrix; b f , b i , b c , b o is the bias term; σ(·) is the Sigmoid activation function, tanh(·) is the hyperbolic tangent activation function; ⊙ represents element-wise multiplication.

[0075] Based on the above, the present invention also provides a crane operation early warning method, which is applied with the above-mentioned crane slewing bearing operation monitoring method.

[0076] Based on the above, the present invention also provides a crane slewing bearing operation monitoring system, which includes:

[0077] A data construction unit is used to collect the acoustic wave data of the slewing bearing of the crane under different working conditions and its corresponding working parameters, and then annotate the acoustic wave data according to the working conditions, and then associate the acoustic wave data and its corresponding working parameters and information annotations to form an acoustic wave data group;

[0078] A data augmentation and collection unit is used to obtain an acoustic wave data set, perform an amplification process on it according to preset conditions, obtain an amplified data set, and then collect the acoustic wave data set and the amplified data set to form an acoustic wave data set;

[0079] A data detection unit is used to extract data from the acoustic wave data set, divide it into training data and verification data, and then input the training data into the neural network for training, and then verify the trained neural network through the verification data until the model converges, and obtain a detection neural network for abnormal detection of the crane slewing bearing;

[0080] The working monitoring unit is used to respond to the working start signal of the crane, monitor the working parameters of the slewing bearing of the crane in real time, collect the acoustic wave data of the slewing bearing in real time, and then compile it into monitoring information;

[0081] A data scheduling unit, used for inputting monitoring information into a detection neural network, and the detection neural network outputs a detection result;

[0082] The information generating unit is used to obtain the detection result, and when the detection result indicates that the slewing bearing of the crane is in an abnormal condition, generate and output a warning message.

[0083] Based on the above, the present invention also provides a computer-readable storage medium, wherein the storage medium stores at least one instruction, at least one program, code set or instruction set, and the at least one instruction, at least one program, code set or instruction set is loaded and executed by a processor to implement the above-mentioned crane slewing bearing operation monitoring method or the above-mentioned crane operation early warning method.

[0084] By adopting the above-mentioned technical scheme, the present invention has the following beneficial effects compared with the prior art: the present scheme is cleverly based on the acoustic wave data group, and by performing data augmentation processing on it, an amplified data group having a certain correlation with the acoustic wave data of the acoustic wave data group is obtained, so that in the subsequent neural network training, the trained detection neural network can have better detection reliability for the acoustic wave data of the slewing bearing; and on this basis, when the present scheme applies the detection neural network to detect the monitoring information of the crane, after performing signal denoising, time-frequency analysis and deep learning network processing on the collected acoustic wave signals, it is possible to obtain acoustic wave data with less noise and better feature signal authenticity, thereby providing better high-quality input data for the reliability application of the detection neural network. BRIEF DESCRIPTION OF THE DRAWINGS

[0085] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0086] Figure 1 This is a brief implementation flow chart of the monitoring method of this program;

[0087] Figure 2 This is a brief unit module connection diagram of the monitoring system of this scheme. DETAILED DESCRIPTION

[0088] The present invention will be further described in detail below in conjunction with the accompanying drawings and examples. It is particularly noted that the following examples are only used to illustrate the present invention, but are not intended to limit the scope of the present invention. Similarly, the following examples are only partial embodiments of the present invention rather than all embodiments, and all other embodiments obtained by those of ordinary skill in the art without creative work are within the scope of protection of the present invention.

[0089] like Figure 1 As shown, the present invention provides a crane slewing bearing operation monitoring method, which comprises:

[0090] S01, collecting the acoustic wave data of the slewing bearing of the crane under different working conditions and its corresponding working parameters, then annotating the acoustic wave data according to the working conditions, and then associating the acoustic wave data and its corresponding working parameters and information annotations to form an acoustic wave data group;

[0091] S02, obtaining an acoustic wave data set, amplifying it according to preset conditions to obtain an amplified data set, and then combining the acoustic wave data set and the amplified data set to form an acoustic wave data set;

[0092] S03, extracting data from the acoustic wave data set, dividing the data into training data and verification data, and then inputting the training data into the neural network for training, and then verifying the trained neural network through the verification data until the model converges, thereby obtaining a detection neural network for abnormality detection of the crane slewing bearing;

[0093] S04, responding to the work start signal of the crane, monitoring the working parameters of the slewing bearing of the crane in real time, and collecting the acoustic wave data of the slewing bearing when it is working in real time, and then compiling it to form monitoring information;

[0094] S05, inputting the monitoring information into the detection neural network, and the detection neural network outputs the detection result;

[0095] S06. Obtain the detection result. When the detection result indicates that the slewing bearing of the crane is in an abnormal condition, generate and output a warning message.

[0096] In terms of the data acquisition source, as a possible implementation mode, further, in the present solution S01, the source of the collected sound wave data is at least one of the following:

[0097] (1) Acoustic wave data collected from the crane's slewing bearing during actual operation;

[0098] (2) constructing a slewing bearing working platform based on the laboratory, making the slewing bearing reach a preset working state according to preset working parameters, and then collecting data to obtain acoustic wave data;

[0099] Regarding information marking, the content of the information marking described in this scheme includes whether the working condition of the slewing bearing is abnormal and the type of abnormality.

[0100] As for the working parameters, the working parameters described in this scheme include: the rotation speed and rotation amplitude of the slewing bearing and the working environment temperature.

[0101] In order to increase the volume of training data and to provide more relevant training data for characteristic signals to improve the training reliability of the detection neural network, as a better implementation option, preferably, in this solution S02, the method for amplifying the acoustic wave data set includes time domain data augmentation processing, frequency domain data augmentation processing, hybrid data augmentation processing or augmentation processing through a deep learning network;

[0102] After the acoustic wave data set is amplified, the working parameters and information annotations of the amplified data set obtained are consistent with the working parameters and information annotations of the corresponding acoustic wave data set; the amplified data set also has an annotation marking it as amplified data.

[0103] As a preferred implementation option for time domain data augmentation processing, preferably, the time domain data augmentation processing described in this solution includes time expansion processing or time shift processing of the sound wave data;

[0104] Time stretching is to amplify the data by changing the time scale of the sound wave data, that is, to obtain the amplified data by compressing or stretching the time axis of the signal without changing the frequency characteristics of the sound wave signal, which includes: using linear interpolation or phase vocoder to adjust the scale factor α of the time axis, and its formula is defined as follows:

[0105]

[0106] Among them, x new (t) represents the sound wave data after time expansion and contraction processing, x(t) represents the initial sound wave data, when α is greater than 1, it represents time expansion, and when α is less than 1, it represents time compression;

[0107] Time shift processing is to translate the sound wave data on the time axis, increase diversity while keeping the frequency characteristics of the signal unchanged, by translating the entire signal forward or backward by a certain time offset δt, which is defined as follows:

[0108] x new (t) = x(t+δt)

[0109] Among them, x new (t) represents the sound wave data after time shift processing, x(t) represents the initial sound wave data, δt represents the translation time, and a negative value represents a backward translation, and a positive value represents a forward translation.

[0110] As a preferred implementation option, the frequency domain data augmentation processing described in this solution preferably includes adding frequency disturbance to the sound wave signal in the frequency domain and performing spectrum trimming on the sound wave signal;

[0111] Frequency perturbation is added to the sound wave signal in the frequency domain to ensure that the diversity of the data is enhanced while maintaining the main frequency characteristics. The frequency domain characteristics are perturbed by adding noise or fine-tuning the spectrum. The formula is defined as follows:

[0112] x new (ω)=X(ω)+ε(ω)

[0113] Among them, ε(ω) is the frequency domain disturbance noise, X(ω) is the sound wave data, and x new (ω) is the acoustic wave data after adding frequency disturbance;

[0114] The spectrum trimming of the sound wave signal is to retain the main part of the signal spectrum, remove or attenuate the secondary frequency components, so as to generate a variety of frequency domain samples; by windowing the signal spectrum, it only retains the signal within a certain frequency range. The formula is defined as follows:

[0115] x new (ω)=X(ω)·W(ω)

[0116] Among them, W(ω) is the window function, X(ω) is the sound wave data, x new (ω) is the sound wave data after spectrum trimming.

[0117] As a preferred implementation option for mixed data augmentation processing, preferably, the mixed data augmentation processing described in this solution includes additive noise processing or mixed sample processing;

[0118] Additive noise processing is to increase the diversity of the data set by adding a small amount of noise to the original signal. It includes adding noise n(t) in the time domain or frequency domain of the sound wave data. The formula is defined as follows:

[0119] x new (t) = x(t) + n(t)

[0120] Among them, x new (t) is the sound wave data after adding noise, x(t) is the sound wave data, and n(t) is Gaussian noise or uniform noise;

[0121] The mixed sample processing is to mix two different samples to generate a new sample to enhance the robustness of the model to different signal modes, which is defined as follows:

[0122] x new(t) = λx1(t) + (1-λ)x2(t)

[0123] Among them, x new (t) is the mixed sound wave data, x1(t) and x2(t) are different sound wave data, and λ is a random number between [0, 1], which is used to balance the mixing ratio of the two signals;

[0124] For augmentation processing through a deep learning network, the method for augmentation processing through a deep learning network described in this scheme includes augmenting samples through a generative adversarial network or augmenting samples through an autoencoder and a variational autoencoder;

[0125] Augmenting samples through generative adversarial networks is to build GANs containing generators and discriminators, and then generate adversarial training between them and the discriminator to generate augmented samples that meet the requirements;

[0126] Augmenting samples through autoencoders and variational autoencoders is the encoding and decoding of original data to generate augmented samples with similar characteristics to the original samples. It includes training the autoencoder network, then compressing the data into the latent space, and reconstructing the decoder to generate new samples. The formula is defined as follows:

[0127] z=f encoder (x)

[0128] x new =f encoder (z+∈)

[0129] Among them, z is the implicit space representation, ∈ is the added random noise, and x new is the new sample generated.

[0130] This solution is cleverly based on the acoustic wave data set. By performing data augmentation processing on it, an amplified data set that has a certain correlation with the acoustic wave data of the acoustic wave data set is obtained, so that when the neural network is subsequently trained, the trained detection neural network can have better detection reliability for the acoustic wave data of the slewing bearing.

[0131] When the crane is working, in order to highlight the authenticity of the sound wave signal and avoid excessive interference caused by noise in the data collection of the slewing bearing, as a better implementation option, preferably, in this solution S04, when collecting the sound wave data of the slewing bearing when working, the collected signal is also subjected to signal denoising, time-frequency analysis and deep learning network processing to remove interference in the signal, which includes:

[0132] (1) Signal denoising

[0133] Use a bandpass filter to denoise the collected acoustic wave signal to remove low-frequency and high-frequency interference signals so that the signal is concentrated in the frequency band containing fault information; assuming that the original signal is x(t) and the filtered signal is y(t), it can be expressed as:

[0134] y(t)=x(t)*h(t)

[0135] Where h(t) is the impulse response function of the filter, and * represents the convolution function;

[0136] (2) Time-frequency analysis

[0137] The short-time Fourier transform (STFT) is used to transform the signal from the time domain to the time-frequency domain to capture the transient characteristics of the signal. The formula is defined as follows:

[0138]

[0139] Among them, X(t,ω) is the complex representation in the time-frequency domain, which is the STFT result, that is, the complex value representation of the signal at time t and frequency ω, which represents the amplitude and phase information of the signal at ω within a small time window at time t; x(τ) is the time domain representation of the original signal x(t), τ represents the time variable, ω(τ-t) is the time window function, which weights the signal within a period of time near time t, τ in the time window function represents the time variable, ω is the frequency variable, e -jωτ is a complex exponential function, which is the core part of Fourier transform and is used to convert signals from time domain to frequency domain. j in the complex exponential function is an imaginary unit, which represents the imaginary part of a complex number. dτ is an integral variable, which represents the integration of time τ. The integration is to accumulate the frequency components of the signal weighted by the window function on the entire time axis.

[0140] (3) Convolutional Neural Network (CNN) processing

[0141] The spatial features of time-frequency data are extracted through the convolutional neural network CNN. The time-frequency feature matrix X(t,ω) obtained by time-frequency analysis is used as input. The output of the lth convolutional layer of the convolutional neural network CNN can be expressed as:

[0142]

[0143] in, is the output feature at position (i, j); is the weight matrix of the convolution kernel, the size of the convolution kernel is k×k; b l is the bias term, σ(·) is the activation function;

[0144] (4) Long Short-Term Memory Network (LSTM) Processing

[0145] The convolution kernel of the convolutional neural network CNN is used to extract features layer by layer, and the extracted features are used as input items of the long short-term memory network LSTM; the long short-term memory network LSTM is used to capture the time dependency of the data, which includes the following calculation steps:

[0146] Forget gate calculation:

[0147] f t =σ(W f ·[h t-1 , x t ]+b f )

[0148] Input gate calculation:

[0149] i t =σ(W i ·[h t-1 , x t ]+b i )

[0150] Candidate memory cell state calculation:

[0151]

[0152] Memory unit status update:

[0153]

[0154] Output gate calculation and final output:

[0155] o t =σ(W o ·[h t-1 , x t ]+b o )

[0156] h t =o t ⊙tanh(C t )

[0157] Among them, x t is the input feature sequence of the long short-term memory network LSTM; h t is the output state of the long short-term memory network LSTM; W f , W i , W C , W o is the weight matrix; b f , b i , b c , b o is the bias term; σ(·) is the Sigmoid activation function, tanh(·) is the hyperbolic tangent activation function; ⊙ represents element-wise multiplication.

[0158] Based on the above, the crane slewing bearing operation monitoring method of this scheme can be applied to the crane operation early warning method to provide reliable protection for the abnormal early warning of the crane.

[0159] Combination Figure 2 As shown, based on the above, this solution also provides a crane slewing bearing working monitoring system, which includes:

[0160] A data construction unit is used to collect the acoustic wave data of the slewing bearing of the crane under different working conditions and its corresponding working parameters, and then annotate the acoustic wave data according to the working conditions, and then associate the acoustic wave data and its corresponding working parameters and information annotations to form an acoustic wave data group;

[0161] A data augmentation and collection unit is used to obtain an acoustic wave data set, perform an amplification process on it according to preset conditions, obtain an amplified data set, and then collect the acoustic wave data set and the amplified data set to form an acoustic wave data set;

[0162] A data detection unit is used to extract data from the acoustic wave data set, divide it into training data and verification data, and then input the training data into the neural network for training, and then verify the trained neural network through the verification data until the model converges, and obtain a detection neural network for abnormal detection of the crane slewing bearing;

[0163] The working monitoring unit is used to respond to the working start signal of the crane, monitor the working parameters of the slewing bearing of the crane in real time, collect the acoustic wave data of the slewing bearing in real time, and then compile it into monitoring information;

[0164] A data scheduling unit, used for inputting monitoring information into a detection neural network, and the detection neural network outputs a detection result;

[0165] The information generating unit is used to obtain the detection result, and when the detection result indicates that the slewing bearing of the crane is in an abnormal condition, generate and output a warning message.

[0166] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0167] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, server, or network device, etc.) or a processor (processor) to perform all or part of the steps of each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.

[0168] The above descriptions are only some embodiments of the present invention, and are not intended to limit the protection scope of the present invention. Any equivalent device or equivalent process transformation made using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A method for monitoring the operation of a slewing bearing of a crane, characterized in that: It includes: S01, collecting the acoustic wave data of the slewing bearing of the crane under different working conditions and its corresponding working parameters, then annotating the acoustic wave data according to the working conditions, and then associating the acoustic wave data and its corresponding working parameters and information annotations to form an acoustic wave data group; S02, obtaining an acoustic wave data set, amplifying it according to preset conditions to obtain an amplified data set, and then combining the acoustic wave data set and the amplified data set to form an acoustic wave data set; S03, extracting data from the acoustic wave data set, dividing the data into training data and verification data, and then inputting the training data into the neural network for training, and then verifying the trained neural network through the verification data until the model converges, thereby obtaining a detection neural network for abnormal detection of the crane slewing bearing; S04, responding to the work start signal of the crane, monitoring the working parameters of the slewing bearing of the crane in real time, and collecting the acoustic wave data of the slewing bearing when it is working in real time, and then compiling it to form monitoring information; S05, inputting the monitoring information into the detection neural network, and the detection neural network outputs the detection result; S06. Obtain the detection result. When the detection result indicates that the slewing bearing of the crane is in an abnormal condition, generate and output a warning message.

2. The crane slewing bearing operation monitoring method according to claim 1, characterized in that: In S01, the source of the collected sound wave data is at least one of the following: (1) Acoustic wave data collected from the crane's slewing bearing during actual operation; (2) constructing a slewing bearing working platform based on the laboratory, making the slewing bearing reach a preset working state according to preset working parameters, and then collecting data to obtain acoustic wave data; The information marking includes whether the slewing bearing working condition is abnormal and the type of abnormality; The working parameters include: the slewing speed, the rotation amplitude and the working environment temperature of the slewing bearing.

3. The crane slewing bearing operation monitoring method according to claim 1 or 2, characterized in that: In S02, the method for amplifying the acoustic wave data set includes time domain data augmentation processing, frequency domain data augmentation processing, mixed data augmentation processing or augmentation processing through a deep learning network; After the acoustic wave data set is amplified, the working parameters and information annotations of the amplified data set obtained are consistent with the working parameters and information annotations of the corresponding acoustic wave data set; the amplified data set also has an annotation marking it as amplified data.

4. The crane slewing bearing operation monitoring method according to claim 3, characterized in that: The time domain data augmentation processing includes time expansion processing or time shift processing on the sound wave data; Time stretching is to amplify the data by changing the time scale of the sound wave data, that is, to obtain the amplified data by compressing or stretching the time axis of the signal without changing the frequency characteristics of the sound wave signal, which includes: using linear interpolation or phase vocoder to adjust the scale factor α of the time axis, and its formula is defined as follows: Among them, x new (t) represents the sound wave data after time expansion and contraction processing, x(t) represents the initial sound wave data, when α is greater than 1, it represents time expansion, and when α is less than 1, it represents time compression; Time shift processing is to translate the sound wave data on the time axis, increase diversity while keeping the frequency characteristics of the signal unchanged, by translating the entire signal forward or backward by a certain time offset δt, which is defined as follows: x new (t)=x(t+δt) Among them, x new (t) represents the sound wave data after time shift processing, x(t) represents the initial sound wave data, δt represents the translation time, and a negative value represents a backward translation, and a positive value represents a forward translation.

5. The crane slewing bearing operation monitoring method according to claim 3, characterized in that: The frequency domain data augmentation processing includes adding frequency disturbance to the sound wave signal in the frequency domain and performing spectrum trimming on the sound wave signal; Frequency perturbation is added to the sound wave signal in the frequency domain to ensure that the diversity of the data is enhanced while maintaining the main frequency characteristics. The frequency domain characteristics are perturbed by adding noise or fine-tuning the spectrum. The formula is defined as follows: x new (ω)=X(ω)+ε(ω) Among them, ε(ω) is the frequency domain disturbance noise, X(ω) is the sound wave data, and x new (ω) is the acoustic wave data after adding frequency disturbance; The spectrum trimming of the sound wave signal is to retain the main part of the signal spectrum, remove or attenuate the secondary frequency components, so as to generate a variety of frequency domain samples; by windowing the signal spectrum, it only retains the signal within a certain frequency range. The formula is defined as follows: x new (ω)=X(ω)·W(ω) Among them, W(ω) is the window function, X(ω) is the sound wave data, x new (ω) is the sound wave data after spectrum trimming.

6. The crane slewing bearing operation monitoring method according to claim 3, characterized in that: The mixed data augmentation process includes additive noise processing or mixed sample processing; The method for augmenting processing by a deep learning network includes augmenting samples by a generative adversarial network or augmenting samples by an autoencoder and a variational autoencoder; Additive noise processing is to increase the diversity of the data set by adding a small amount of noise to the original signal. It includes adding noise n(t) in the time domain or frequency domain of the sound wave data. The formula is defined as follows: X new (t)=x(t)+n(t) Among them, x new (t) is the sound wave data after adding noise, x(t) is the sound wave data, and n(t) is Gaussian noise or uniform noise; The mixed sample processing is to mix two different samples to generate a new sample to enhance the robustness of the model to different signal modes, which is defined as follows: x new (t)=λx1(t)+(1-λ)x2(t) Among them, x new (t) is the mixed sound wave data, x1(t) and x2(t) are different sound wave data, and λ is a random number between [0, 1], which is used to balance the mixing ratio of the two signals; Augmenting samples through generative adversarial networks is to build GANs containing generators and discriminators, and then generate adversarial training between them and the discriminator to generate augmented samples that meet the requirements; Augmenting samples through autoencoders and variational autoencoders is the encoding and decoding of original data to generate augmented samples with similar characteristics to the original samples. It includes training the autoencoder network, then compressing the data into the latent space, and reconstructing the decoder to generate new samples. The formula is defined as follows: z=f encoder (x) x new =f encoder (z+∈) Among them, z is the implicit space representation, ∈ is the added random noise, and x new is the new sample generated.

7. The crane slewing bearing operation monitoring method according to claim 3, characterized in that: In S04, when collecting the acoustic wave data of the slewing bearing in operation, the collected signals are also subjected to signal denoising, time-frequency analysis and deep learning network processing to remove interference in the signals, which includes: (1) Signal denoising Use a bandpass filter to denoise the collected acoustic wave signal to remove low-frequency and high-frequency interference signals so that the signal is concentrated in the frequency band containing fault information; assuming that the original signal is x(t) and the filtered signal is y(t), it can be expressed as: y(t)=x(t)*h(t) Where h(h) is the impulse response function of the filter, and * represents the convolution function; (2) Time-frequency analysis The short-time Fourier transform (STFT) is used to transform the signal from the time domain to the time-frequency domain to capture the transient characteristics of the signal. The formula is defined as follows: Among them, X(t,ω) is the complex representation in the time-frequency domain, which is the STFT result, that is, the complex value representation of the signal at time t and frequency ω, which represents the amplitude and phase information of the signal at ω within a small time window at time t; x(t) is the time domain representation of the original signal x(t), τ represents the time variable, ω(τ-t) is the time window function, which weights the signal within a period of time near time t, τ in the time window function represents the time variable, ω is the frequency variable, e -jωτ is a complex exponential function, which is the core part of Fourier transform and is used to convert signals from time domain to frequency domain. j in the complex exponential function is an imaginary unit, which represents the imaginary part of a complex number. dτ is an integral variable, which represents the integration of time τ. The integration is to accumulate the frequency components of the signal weighted by the window function on the entire time axis. (3) Convolutional Neural Network (CNN) processing The spatial features of time-frequency data are extracted through the convolutional neural network CNN. The time-frequency feature matrix X(t,ω) obtained by time-frequency analysis is used as input. The output of the lth convolutional layer of the convolutional neural network CNN can be expressed as: in, is the output feature at position (i, j); is the weight matrix of the convolution kernel, the size of the convolution kernel is k×k; b l is the bias term, σ(·) is the activation function; (4) Long Short-Term Memory Network (LSTM) Processing The convolution kernel of the convolutional neural network CNN is used to extract features layer by layer, and the extracted features are used as input items of the long short-term memory network LSTM; the long short-term memory network LSTM is used to capture the time dependency of the data, which includes the following calculation steps: Forget gate calculation: f t =σ(W f ·[h t-1 ,x t ]+b f ) Input gate calculation: i t =σ(W i ·[h t-1 ,x t ]+b i Candidate memory cell state calculation: Memory unit status update: Output gate calculation and final output: the t =σ(W o ·[h t-1 ,x t ]+b o ) h t =o t ⊙tanh(C t ) Among them, x t is the input feature sequence of the long short-term memory network LSTM; h t is the output state of the long short-term memory network LSTM; W f , W i , W C , W o is the weight matrix; b f 、b i 、b c 、b o is the bias term; σ(·) is the Sigmoid activation function, tanh(·) is the hyperbolic tangent activation function; ⊙ represents element-wise multiplication.

8. A crane operation early warning method, characterized in that: The invention is applied with the crane slewing bearing operation monitoring method as claimed in any one of claims 1 to 7.

9. A crane slewing bearing operation monitoring system, characterized in that: It includes: A data construction unit is used to collect the acoustic wave data of the slewing bearing of the crane under different working conditions and its corresponding working parameters, and then annotate the acoustic wave data according to the working conditions, and then associate the acoustic wave data and its corresponding working parameters and information annotations to form an acoustic wave data group; A data augmentation and collection unit is used to obtain an acoustic wave data set, perform an amplification process on it according to preset conditions, obtain an amplified data set, and then collect the acoustic wave data set and the amplified data set to form an acoustic wave data set; A data detection unit is used to extract data from the acoustic wave data set, divide it into training data and verification data, and then input the training data into the neural network for training, and then verify the trained neural network through the verification data until the model converges, and obtain a detection neural network for abnormal detection of the crane slewing bearing; The working monitoring unit is used to respond to the working start signal of the crane, monitor the working parameters of the slewing bearing of the crane in real time, collect the acoustic wave data of the slewing bearing in real time, and then compile it into monitoring information; A data scheduling unit, used for inputting monitoring information into a detection neural network, and the detection neural network outputs a detection result; The information generating unit is used to obtain the detection result, and when the detection result indicates that the slewing bearing of the crane is in an abnormal condition, generate and output a warning message.

10. A computer-readable storage medium, characterized in that: The storage medium stores at least one instruction, at least one program, code set or instruction set, and the at least one instruction, at least one program, code set or instruction set is loaded and executed by the processor to implement the crane slewing bearing operation monitoring method as described in one of claims 1 to 7 or the crane operation early warning method as described in claim 8.

Images