A slurry pump fault detection and life prediction system and method based on big data

Abstract

The present invention relates to the technical field of data analysis, and discloses a slurry pump fault detection and life prediction system and method based on big data. The present invention first obtains the slurry pump vibration signal, detects the abnormal vibration signal and deletes the abnormal vibration signal, and then performs noise reduction processing on the slurry pump vibration signal; secondly, the slurry pump vibration signal after noise reduction is processed by segmented aggregation approximation and Gram angle field to generate a slurry pump vibration signal image, and extracts the slurry pump vibration signal image features, trains a convolutional neural network, and optimizes the convolutional neural network parameters based on the electric eel foraging optimization algorithm to obtain a convolutional neural network recognition model to achieve slurry pump fault detection; then, the slurry pump vibration signal IMF component is decomposed, the LSTM neural network is trained, the LSTM neural network prediction model is obtained, and the slurry pump life prediction value is output. The present invention achieves the purpose of fault detection and life prediction by processing the slurry pump vibration signal, and the method is accurate and objective.

Description

Technical Field

[0001] The present invention relates to the technical field of data analysis, and in particular to a slurry pump fault detection and life prediction system and method based on big data. Background Art

[0002] Traditional slurry pump fault detection judges the slurry pump's operating sound and previous service life based on experience, which has problems such as large errors and slow detection speed. At the same time, the traditional slurry pump life prediction method judges the degree of wear of the slurry pump. The hardness of the pump conveying medium and solid materials is often difficult to measure. Different slurry pumps are made of different materials, which inevitably leads to errors in the slurry pump life prediction results. Moreover, slurry pump fault detection and life prediction do not use high-tech technologies such as neural networks, which brings inconvenience to data analysis. Summary of the Invention

[0003] In response to the problems in the related art, the present invention provides a slurry pump fault detection and life prediction system and method based on big data to overcome the above-mentioned technical problems existing in the existing related art.

[0004] To solve the above technical problems, the present invention is achieved through the following technical solutions:

[0005] The present invention is a method for slurry pump fault detection and life prediction based on big data, comprising the following steps:

[0006] S1. Collect the vibration signal of the slurry pump in the running state to obtain the slurry pump vibration signal, perform abnormal vibration signal detection on the slurry pump vibration signal based on the outlier factor algorithm, and delete the abnormal vibration signal; then perform singular value decomposition and reconstruction signal processing to complete the slurry pump vibration signal noise reduction and obtain the noise-reduced slurry pump vibration signal;

[0007] S2, the slurry pump vibration signal after noise reduction is formed into a slurry pump vibration signal set, and then the segmented aggregation approximation and Gram angle field processing are performed to generate a slurry pump vibration signal image, and then feature extraction is performed based on the Gauss-Markov random field to obtain a 12-dimensional feature vector;

[0008] S3. Collect new slurry pump vibration signals to obtain new 12-dimensional feature vectors. Use the generator and discriminator to augment the new 12-dimensional feature vectors. Then, train the convolutional neural network and optimize the convolutional neural network parameters based on the electric eel foraging optimization algorithm to obtain a convolutional neural network recognition model to implement slurry pump fault detection.

[0009] S4. Obtain the original vibration signal of the slurry pump, obtain the IMF component of the original vibration signal of the slurry pump based on the empirical mode decomposition method, and train the LSTM neural network to obtain the LSTM neural network prediction model. Combined with the slurry pump vibration signal, the predicted value of the slurry pump service life is obtained to complete the slurry pump life prediction.

[0010] The invention first obtains the slurry pump vibration signal by collecting the vibration signal of the slurry pump in the running state, performs abnormal vibration signal detection on the slurry pump vibration signal, deletes the abnormal vibration signal, and then uses singular value decomposition and reconstruction signal to retain the effective information in the slurry pump vibration signal to achieve noise reduction of the slurry pump vibration signal; secondly, the slurry pump vibration signal after noise reduction is processed by segmented aggregation approximation and Gram angle field to generate a slurry pump vibration signal image. By setting the time series, the calculation complexity can be reduced and the signal quality can be improved. Then feature extraction is performed, and the extracted 12-dimensional feature vector effectively reduces redundancy and improves the accuracy of subsequent data processing; and then the new slurry pump vibration is used. The convolutional neural network is trained based on signal extraction features, and the generator and discriminator are introduced for augmentation processing to minimize the error between the generated samples and the real samples. The convolutional neural network parameters are optimized based on the electric eel foraging optimization algorithm to obtain a convolutional neural network recognition model, which can determine whether the slurry pump is faulty. The electric eel foraging optimization algorithm finds the optimal solution to the problem by simulating the hunting behavior of electric eels, and has strong optimization ability and fast convergence speed. The IMF component of the original vibration signal of the slurry pump is then extracted, and the LSTM neural network is trained to obtain the LSTM neural network prediction model, and finally the predicted value of the slurry pump service life is output. The LSTM neural network has good prediction accuracy and can realize the life prediction of the slurry pump.

[0011] Preferably, the S1 comprises the following steps:

[0012] S11. When the slurry pump is in operation, a vibration sensor is placed at the center of a bearing of the slurry pump to collect a bearing vibration signal of the slurry pump, the bearing vibration signal of the slurry pump is converted into a 485 signal, and the 485 signal is regarded as a slurry pump vibration signal; a slurry pump vibration signal collection time interval is set, and the slurry pump vibration signal is repeatedly collected at an α sampling frequency to obtain a slurry pump vibration signal set, wherein the slurry pump vibration signal set includes a normal slurry pump vibration signal, a single fault slurry pump vibration signal, and a compound fault slurry pump vibration signal;

[0013] S12, set the sliding window size to 100, perform sliding sampling in the slurry pump vibration signal set with a step size of 100, generate a slurry pump vibration signal data segment with a length of 100 in the slurry pump vibration signal set, and form a slurry pump vibration signal data set A = {b1, b2, b3, ..., b a}, where b arepresents the ath slurry pump vibration signal data; select the ath slurry pump vibration signal data set The vibration signal data of a slurry pump is Data points, calculation The distance size of the other slurry pump vibration signal data in the slurry pump vibration signal data set is arranged in ascending order to obtain a distance size set, and the slurry pump vibration signal data corresponding to the distance less than k distance is found in the distance size set, and the slurry pump vibration signal data corresponding to the distance less than k distance is recorded as k distance Data point neighborhood;

[0014] When the slurry pump vibration signal data in the slurry pump vibration signal data set is within the k distance When the data point is within the neighborhood, it is recorded as data reachable, otherwise the data is unreachable; when the data is reachable, The calculation formula for the reachable data density of a data point is as follows:

[0015]

[0016] in, express The reachable data density of the data point, average represents the average value, Represents k distance The number of data points in the neighborhood of the data point, express distance to k In the neighborhood of a data point The distance between the data points,

[0017] According to the The reachable data density of a data point is calculated The outlier factor of a data point is calculated as follows:

[0018]

[0019] in, express The outlier factor of the data point, B(i1) represents the k distance In the neighborhood of a data point The reachable data density of the data points,

[0020] Calculate the outlier factors of the slurry pump vibration signal data segments in the slurry pump vibration signal data set in sequence, set the outlier threshold as ω, delete the data whose outlier factors are less than ω in the slurry pump vibration signal data segments in the slurry pump vibration signal data set, complete the deletion of abnormal vibration signal data, and obtain the abnormally deleted slurry pump vibration signal set;

[0021] S13, set the abnormal deleted slurry pump vibration signal set to A1 = {b′1, b′2, b′3, ..., b′ a′}, where b′ a′ represents the a′th slurry pump vibration signal, and the dimension of the slurry pump vibration signal in the abnormal deleted slurry pump vibration signal set is set to c, and the slurry pump vibration signal matrix C is generated as follows:

[0022]

[0023] Among them, b′ a′c represents the vibration signal of the slurry pump with the dimension of a′;

[0024] Set the left singular matrix C1, the right singular matrix C2 and the diagonal matrix and satisfy Among them C2 T Represents the transpose of the right singular matrix, and sets the column vector of the left singular matrix to the left singular vector of the slurry pump vibration signal matrix, and the column vector of the right singular matrix to the right singular vector of the slurry pump vibration signal matrix. At this time, the diagonal elements of the slurry pump vibration signal matrix are recorded as the singular value set A2 = {b″1, b″2, b″3, ..., b″ a′}, where b″ a′ Represents the a′th singular value, completing the singular value decomposition; sorting the singular values ​​in the singular value set in descending order, setting the singular value threshold to ξ, discarding the singular values ​​in the singular value set that are less than ξ, and forming a valid singular value set from the singular value set that are greater than or equal to ξ, denoted as A3 = {b″′1, b″′2, b″′3, ..., b″′ a″}, where b″′ a″ Represents the a″th valid singular value, and the signal components corresponding to the valid singular values ​​in the valid singular value set are superimposed to obtain the reconstructed slurry pump vibration signal matrix, and the slurry pump vibration signal noise reduction is completed. The calculation formula of the slurry pump vibration signal after noise reduction is as follows:

[0025]

[0026] Among them, B″ represents the vibration signal of the slurry pump after noise reduction, represents the i3th left singular vector of the reconstructed slurry pump vibration signal matrix, represents the transpose of the i3th right singular vector of the reconstructed slurry pump vibration signal matrix, represents the i3th valid singular value on the diagonal of the diagonal matrix, i3 = 1, 2, 3, ..., a″.

[0027] This invention obtains the vibration signal of the slurry pump by collecting the vibration signal of the slurry pump in the running state, performs abnormal vibration signal detection on the slurry pump vibration signal based on the LOF algorithm, deletes the abnormal vibration signal, and then uses singular value decomposition and reconstructs the signal. By superimposing the signal components, the effective information in the slurry pump vibration signal is retained to achieve noise reduction of the slurry pump vibration signal.

[0028] Preferably, said S2 comprises the following steps:

[0029] S21, forming a noise-reduced slurry pump vibration signal set according to the noise-reduced slurry pump vibration signal, denoted as A4={d1, d2, d3, ..., d e}, where d e represents the e-th noise-reduced slurry pump vibration signal, collects the time corresponding to the noise-reduced slurry pump vibration signal in the noise-reduced slurry pump vibration signal set, generates a slurry pump vibration signal time series, and divides the slurry pump vibration signal time series into a group of e′ slurry pump vibration signal time series. a slurry pump vibration signal time subsequence, and subjecting the slurry pump vibration signal time subsequence to segmented aggregation approximation processing to obtain a processed slurry pump vibration signal time series;

[0030] S22, compressing the processed slurry pump vibration signal time series to the interval [-1, 1] to obtain a compressed slurry pump vibration signal time series A5, mapping the compression value in the compressed slurry pump vibration signal time series into an angle χ1, setting the time mapping radius to e1, and calculating the formula as follows:

[0031]

[0032] Where t represents the time corresponding to the processed slurry pump vibration signal time series, and δ represents the polar coordinate span coefficient;

[0033] The compressed values ​​in the compressed slurry pump vibration signal time series are again mapped into an angle χ2, and the Gram matrix C1=cos(χ1+χ2); in the Gram matrix, each data element represents the pixel value of a pixel point of the slurry pump vibration signal image, thereby generating a slurry pump vibration signal image;

[0034] S23, extracting texture features from the slurry pump vibration signal image using a Gauss-Markov random field, the specific steps are as follows:

[0035] S231, select pixel points on the slurry pump vibration signal image to obtain a pixel point set, set Represents the i4th pixel in the pixel set, represents the weight of the pixel point in the symmetric Gauss-Markov neighborhood, Represents Gaussian distribution noise, then the Gauss-Markov model calculation formula is as follows,

[0036]

[0037] Where D represents the Gauss-Markov model, g represents the number of pixel points, and i4 = 1, 2, 3, ..., g;

[0038] S232, substituting the pixel points of the slurry pump vibration signal image into the Gauss-Markov model to obtain a differential equation. The calculation formula is as follows:

[0039]

[0040] Where D′ represents the difference equation, represents the eigenvector, C2 represents the difference equation matrix;

[0041] The difference equation error is solved using a least square error method, a 12-dimensional vector that satisfies the difference equation error is selected, and the 12-dimensional vector that satisfies the difference equation error is recorded as a 12-dimensional eigenvector.

[0042] This invention generates a slurry pump vibration signal image by using segmented aggregation approximation and Gram angle field processing on the noise-reduced slurry pump vibration signal. By setting the time series, the computational complexity can be reduced and the signal quality can be improved. Feature extraction is then performed based on the Gauss-Markov random field to obtain a 12-dimensional feature vector, which can effectively reduce redundancy and improve data processing accuracy.

[0043] Preferably, the step S3 includes the following steps:

[0044] S31. The deep convolutional generative adversarial network augmentation includes a generator and a discriminator. The generator consists of a reshape layer and a deconvolution layer. The activation function is set to the ReLU function. The data is input into the deep convolutional generative adversarial network through the reshape layer. The reshape layer reshapes the input data sample and then the deconvolution layer amplifies the feature to generate sample similarity features; the discriminator consists of a convolution layer and a fully connected layer. The input data sample and the sample similarity feature are subjected to feature extraction and dimensionality reduction processing. The sigmoid function is used as the activation function and the result is output after passing through the fully connected layer; the generator and the discriminator are alternately trained on the data samples until Nash equilibrium is reached to obtain augmented sample data;

[0045] A new slurry pump vibration signal is collected to obtain a new slurry pump vibration signal set. After outlier deletion and noise reduction processing are performed on the new slurry pump vibration signal set, a new slurry pump vibration signal image is generated to obtain a new 12-dimensional feature vector. The new 12-dimensional feature vector is divided into a feature training set and a feature test set. The feature training set is input into a generator to generate vibration signal similarity features. The vibration signal similarity features are then input into a discriminator to determine whether Nash equilibrium is reached. If Nash equilibrium is reached, the discriminator outputs an augmented feature training set. Otherwise, S31 is repeated until Nash equilibrium is reached.

[0046] S32, inputting the augmented feature training set into the convolutional neural network, setting the current number of iterations to g~ and the maximum number of iterations to G, and stopping the iteration when the current number of iterations reaches the maximum number of iterations to obtain a trained convolutional neural network; inputting the feature test set into the trained convolutional neural network, and optimizing the convolution kernel size and step size using the electric eel foraging optimization algorithm to obtain a convolutional neural network recognition model, the specific steps are as follows:

[0047] S321, using the trained convolutional neural network recognition accuracy as the fitness function, the electric eel population in the interaction stage, using stirring as the interaction sign in the search space, setting the current iteration number to h, the maximum iteration number to H, selecting a random electric eel j from the electric eel population, and setting the random electric eel h-th iteration position x j (h), the position of the i-th electric eel in the electric eel population at the h-th iteration The average position of the electric eels in the electric eel population at the hth iteration is recorded as ε represents the interaction parameter, then the position of the i-th electric eel in the electric eel population at the h+1th iteration is The calculation formula is as follows,

[0048] When the fitness function value of the random electric eel in the electric eel population is less than the fitness function value of the i-th electric eel in the electric eel population,

[0049] When the fitness function value of a random electric eel in the electric eel population is greater than or equal to the fitness function value of the i-th electric eel in the electric eel population,

[0050] When the electric eel population is in the resting stage, the position vector of the electric eel in the electric eel population is projected onto the main diagonal of the search space to establish a rest area; a three-dimensional coordinate axis is established for normalization, and the search space and the eel position are stipulated to be within the range of 0-1 on the three-dimensional coordinate axis. The projected position is regarded as the rest area, and the electric eel position corresponding to the best fitness function value of the hth iteration is set as The initial scale of the rest area is φ, and the scale of the rest area is The upper bound of the rest area is k′, the upper bound of the rest area is k″, the normalized number of the hth iteration is E(h), g1 represents a random number and g1∈(0,1), then the calculation formula for the h+1th iteration rest position x2(h+1) is as follows,

[0051]

[0052] After the resting position is determined, the electric eel population moves to the resting position. Let g2 represent a random number between 0 and 1 that obeys the normal distribution, g3 represent a random number and g3∈(0,1), round represents rounding, then the resting position of the i-th electric eel in the electric eel population at the h+1th iteration is The calculation formula is as follows,

[0053]

[0054] S322, in the hunting stage, the electric eels surround the prey and continuously accelerate the search for the local fitness function value. The electric eels first determine the pre-position of the hunting area and set the scale of the hunting area to γ1. The h-th iterative hunting position of the electric eels in the electric eel population is recorded as The prey position is x′, and the electric eel hunting range is The electric eel moves to the pre-position of the hunting area, and the pre-position of the hunting area in the hth iteration is set to x4(h). The calculation formula for the pre-position of the hunting area in the h+1th iteration is x5(h+1).

[0055]

[0056] After the pre-position of the hunting area is determined, the electric eel population moves to the pre-position of the hunting area. The curl factor is set to η, g4 represents a random number and g4∈(0,1), then the pre-position of the hunting area of ​​the i-th electric eel in the electric eel population in the h+1th iteration is The calculation formula is as follows,

[0057]

[0058] The electric eel population moves to the pre-position of the hunting area, and the electric eel population begins the migration phase. At this time, the optimal fitness function value is continuously sought, and the position of the h+1th iteration hunting area of ​​the i-th electric eel in the electric eel population is recorded as When the fitness function value of the i-th electric eel in the electric eel population at the h+1 iteration is less than the fitness function value of the i-th electric eel in the electric eel population at the h iteration, the position of the i-th electric eel in the electric eel population at the h+1 iteration is Otherwise, the position of the i-th electric eel in the electric eel population at the h+1th iteration is The electric eel continuously iterates its own position. When the current number of iterations is equal to the maximum number of iterations, the iteration stops and the final position of the i-th electric eel in the electric eel population is obtained. The coordinate values ​​of the two-dimensional coordinates of the final position of the i-th electric eel in the electric eel population correspond to the convolution kernel size and step size respectively;

[0059] S33. Input the 12-dimensional feature vector into the generator and the discriminator to obtain an augmented feature vector, and input the augmented feature vector into the convolutional neural network recognition model. If the convolutional neural network recognition model outputs 0, it indicates that the vibration signal of the slurry pump is abnormal and the slurry pump is faulty. If the convolutional neural network recognition model outputs 1, it indicates that the vibration signal of the slurry pump is normal and the slurry pump is not faulty, thereby completing the slurry pump fault detection.

[0060] This invention trains the convolutional neural network by using new slurry pump vibration signal extraction features, introduces generators and discriminators for augmentation processing, minimizes the error between generated samples and real samples, and optimizes the convolutional neural network parameters based on the electric eel foraging optimization algorithm. By simulating the hunting behavior of electric eels, the optimal solution is found, and a convolutional neural network recognition model is obtained to determine whether the slurry pump is faulty.

[0061] Preferably, said S4 comprises the following steps:

[0062] S41. Reacquire the original vibration signal of the slurry pump, set the upper envelope of the original vibration signal of the slurry pump to l1, set the lower envelope of the original vibration signal of the slurry pump to l2, calculate the average of the upper envelope of the original vibration signal of the slurry pump and the lower envelope of the original vibration signal of the slurry pump, and obtain the average envelope of the vibration signal; use the original vibration signal of the slurry pump to subtract the average envelope of the vibration signal to obtain a new vibration signal component, calculate the average of the upper envelope of the new vibration signal component and the lower envelope of the new vibration signal component, and record it as A new vibration signal average envelope curve; when the new vibration signal average envelope curve is 0, and the absolute value of the difference between the number of zero points and the number of extreme value points of the new vibration signal average envelope curve on the time domain diagram is less than or equal to 1, the new vibration signal component is used as the first IMF (inherent mode function) component; otherwise, S41 is repeated until the new vibration signal average envelope curve is 0, and the absolute value of the difference between the number of zero points and the number of extreme value points of the new vibration signal average envelope curve on the time domain diagram is less than or equal to 1, thereby obtaining the first IMF component;

[0063] S42, removing the first IMF component from the original vibration signal of the slurry pump, decomposing it to obtain a second original vibration signal of the slurry pump, performing S41 on the second original vibration signal of the slurry pump to obtain a second IMF component; continuously repeating S41 until the remaining original vibration signal of the slurry pump is a monotonic function, stopping the decomposition, recording the remaining original vibration signal of the slurry pump as an original residual vibration signal, and outputting the IMF component of the original vibration signal of the slurry pump and the original residual vibration signal;

[0064] In the LSTM (long short-term memory) neural network, the number of IMF components of the original vibration signal of the slurry pump is set to m, the sliding window size is n, the number of neurons is (m+2)×n, the current number of iterations is q, the maximum number of iterations is Q, the IMF components of the original vibration signal of the slurry pump, the original residual vibration signal and the original vibration signal of the slurry pump are used as the prediction sample set, the prediction sample set is divided into a prediction training set and a prediction test set, the prediction training set is input into the LSTM neural network, and when the current number of iterations reaches the maximum number of iterations, the trained LSTM neural network is obtained; the prediction test set is input into the trained LSTM neural network, and the error threshold is set to ψ. When the output error is less than ψ, the LSTM neural network prediction model is obtained, otherwise the weight is adjusted until the output error is less than ψ;

[0065] The IMF component of the slurry pump vibration signal is extracted to obtain the IMF component of the original vibration signal of the slurry pump and the residual vibration signal. The IMF component of the original vibration signal of the slurry pump, the residual vibration signal and the slurry pump vibration signal are used as a prediction verification set, and the prediction verification set is input into the LSTM neural network prediction model to output the predicted value of the slurry pump service life to complete the slurry pump life prediction.

[0066] This invention obtains the original vibration signal of the slurry pump again, extracts the IMF component of the original vibration signal of the slurry pump, and trains the LSTM neural network through the IMF component to obtain an LSTM neural network prediction model. The LSTM neural network has good prediction accuracy and finally outputs the predicted value of the slurry pump service life to achieve slurry pump life prediction.

[0067] This embodiment also discloses a system for slurry pump fault detection and life prediction based on big data, which specifically includes: a slurry pump vibration signal noise reduction module, a slurry pump vibration signal feature extraction module, a slurry pump fault detection module and a slurry pump life prediction module;

[0068] The slurry pump vibration signal noise reduction module is used to detect and delete abnormal vibration signals, and then achieve noise reduction after singular value decomposition and reconstruction signal processing;

[0069] The slurry pump vibration signal feature extraction module is used to generate a slurry pump vibration signal image based on the vibration signal, and obtain a 12-dimensional feature vector after feature extraction;

[0070] The slurry pump fault detection module is used to train a convolutional neural network to obtain a convolutional neural network recognition model to detect slurry pump faults;

[0071] The slurry pump life prediction module is used to train the LSTM neural network to obtain an LSTM neural network prediction model and output a predicted value of the slurry pump service life.

[0072] The present invention has the following beneficial effects:

[0073] 1. The invention obtains the vibration signal of the slurry pump by collecting the vibration signal of the slurry pump in the running state, detects the abnormal vibration signal of the slurry pump vibration signal, deletes the abnormal vibration signal, and then uses singular value decomposition and reconstructs the signal to realize the noise reduction of the slurry pump vibration signal; and uses segmented aggregation approximation and Gram angle field processing on the slurry pump vibration signal after noise reduction to generate a slurry pump vibration signal image, and then performs feature extraction based on Gauss-Markov random field to extract 12-dimensional feature vectors, which can effectively reduce redundancy and improve data processing accuracy.

[0074] 2. This invention uses new slurry pump vibration signal extraction features to train the convolutional neural network, introduces a generator and a discriminator to perform augmentation processing on the 12-dimensional feature vector, which can reduce sample errors, and optimizes the convolutional neural network parameters based on the electric eel foraging optimization algorithm. By simulating the hunting behavior of electric eels, the optimal solution is found, and a convolutional neural network recognition model is obtained to realize slurry pump fault detection.

[0075] 3. The invention obtains the original vibration signal of the slurry pump again, extracts the IMF component of the original vibration signal of the slurry pump, and trains the LSTM neural network to obtain an LSTM neural network prediction model. The LSTM neural network has good prediction accuracy, outputs the predicted value of the service life of the slurry pump, and completes the life prediction of the slurry pump.

[0076] Of course, any product implementing the present invention does not necessarily need to achieve all of the advantages described above at the same time. BRIEF DESCRIPTION OF THE DRAWINGS

[0077] In order to more clearly illustrate the technical solutions of the embodiments of the invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the invention. For ordinary technicians in this field, they can also obtain drawings based on these drawings without paying any creative work.

[0078] Figure 1The present invention provides a flow chart of a slurry pump fault detection and life prediction system based on big data for slurry pump fault detection and life prediction. DETAILED DESCRIPTION

[0079] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0080] In the description of the present invention, it should be understood that the terms "opening", "upper", "lower", "top", "middle", "inside" and the like indicating orientation or positional relationship are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the components or elements referred to must have a specific orientation, be constructed and operate in a specific orientation, and therefore cannot be understood as limiting the invention.

[0081] This paper discloses a method for slurry pump fault detection and life prediction based on big data, which specifically includes the following contents:

[0082] S1. Collect the vibration signal of the slurry pump in the running state to obtain the slurry pump vibration signal, perform abnormal vibration signal detection on the slurry pump vibration signal based on the outlier factor algorithm, and delete the abnormal vibration signal; then perform singular value decomposition and reconstruction signal processing to complete the slurry pump vibration signal noise reduction and obtain the noise-reduced slurry pump vibration signal;

[0083] Said S1 comprises the following steps:

[0084] S11. When the slurry pump is in operation, a vibration sensor is placed at the center of a bearing of the slurry pump to collect a bearing vibration signal of the slurry pump, the bearing vibration signal of the slurry pump is converted into a 485 signal, and the 485 signal is regarded as a slurry pump vibration signal; a slurry pump vibration signal collection time interval is set, and the slurry pump vibration signal is repeatedly collected at an α sampling frequency to obtain a slurry pump vibration signal set, wherein the slurry pump vibration signal set includes a normal slurry pump vibration signal, a single fault slurry pump vibration signal, and a compound fault slurry pump vibration signal;

[0085] S12, set the sliding window size to 100, perform sliding sampling in the slurry pump vibration signal set with a step size of 100, generate a slurry pump vibration signal data segment with a length of 100 in the slurry pump vibration signal set, and form a slurry pump vibration signal data set A = {b1, b2, b3, ..., b a}, where b arepresents the ath slurry pump vibration signal data; select the ath slurry pump vibration signal data set The vibration signal data of a slurry pump is Data points, calculation The distance size of the other slurry pump vibration signal data in the slurry pump vibration signal data set is arranged in ascending order to obtain a distance size set, and the slurry pump vibration signal data corresponding to the distance less than k distance is found in the distance size set, and the slurry pump vibration signal data corresponding to the distance less than k distance is recorded as k distance Data point neighborhood;

[0086] When the slurry pump vibration signal data in the slurry pump vibration signal data set is within the k distance When the data point is within the neighborhood, it is recorded as data reachable, otherwise the data is unreachable; when the data is reachable, The calculation formula for the reachable data density of a data point is as follows:

[0087]

[0088] in, express The reachable data density of the data point, average represents the average value, Represents k distance The number of data points in the neighborhood of the data point, express To k distance In the neighborhood of a data point The distance between the data points,

[0089] According to the The reachable data density of a data point is calculated The outlier factor of a data point is calculated as follows:

[0090]

[0091] in, express The outlier factor of the data point, B(i1) represents the k distance In the neighborhood of a data point The reachable data density of the data points,

[0092] Calculate the outlier factors of the slurry pump vibration signal data segments in the slurry pump vibration signal data set in sequence, set the outlier threshold as ω, delete the data whose outlier factors are less than ω in the slurry pump vibration signal data segments in the slurry pump vibration signal data set, complete the deletion of abnormal vibration signal data, and obtain the abnormally deleted slurry pump vibration signal set;

[0093] S13, set the abnormal deleted slurry pump vibration signal set to A1 = {b′1, b′2, b′3, ..., b′ a′}, where b′ a′ represents the a′th slurry pump vibration signal, and the dimension of the slurry pump vibration signal in the abnormal deleted slurry pump vibration signal set is set to c, and the slurry pump vibration signal matrix C is generated as follows:

[0094]

[0095] Among them, b′ a′c represents the vibration signal of the slurry pump with the dimension of a′;

[0096] Set the left singular matrix C1, the right singular matrix C2 and the diagonal matrix and satisfy Among them C2 T Represents the transpose of the right singular matrix, and sets the column vector of the left singular matrix to the left singular vector of the slurry pump vibration signal matrix, and the column vector of the right singular matrix to the right singular vector of the slurry pump vibration signal matrix. At this time, the diagonal elements of the slurry pump vibration signal matrix are recorded as the singular value set A2 = {b″1, b″2, b″3, ..., b″ a′}, where b″ a′ Represents the a′th singular value, completing the singular value decomposition; sorting the singular values ​​in the singular value set in descending order, setting the singular value threshold to ξ, discarding the singular values ​​in the singular value set that are less than ξ, and forming a valid singular value set from the singular value set that are greater than or equal to ξ, denoted as A3 = {b″′1, b″′2, b″′3, ..., b″′ a″}, where b″′ a″ Represents the a″th valid singular value, and the signal components corresponding to the valid singular values ​​in the valid singular value set are superimposed to obtain the reconstructed slurry pump vibration signal matrix, and the slurry pump vibration signal noise reduction is completed. The calculation formula of the slurry pump vibration signal after noise reduction is as follows:

[0097]

[0098] Among them, B″ represents the vibration signal of the slurry pump after noise reduction, represents the i3th left singular vector of the reconstructed slurry pump vibration signal matrix, represents the transpose of the i3th right singular vector of the reconstructed slurry pump vibration signal matrix, represents the i3th valid singular value on the diagonal of the diagonal matrix, i3=1,2,3,...,a″;

[0099] S2, the slurry pump vibration signal after noise reduction is formed into a slurry pump vibration signal set, and then the segmented aggregation approximation and Gram angle field processing are performed to generate a slurry pump vibration signal image, and then feature extraction is performed based on the Gauss-Markov random field to obtain a 12-dimensional feature vector;

[0100] The S2 comprises the following steps:

[0101] S21, forming a noise-reduced slurry pump vibration signal set according to the noise-reduced slurry pump vibration signal, denoted as A4 = {d1, d2, d3, ..., d e}, where d e represents the e-th noise-reduced slurry pump vibration signal, collects the time corresponding to the noise-reduced slurry pump vibration signal in the noise-reduced slurry pump vibration signal set, generates a slurry pump vibration signal time series, and divides the slurry pump vibration signal time series into a group of e′ slurry pump vibration signal time series. a slurry pump vibration signal time subsequence, and subjecting the slurry pump vibration signal time subsequence to segmented aggregation approximation processing to obtain a processed slurry pump vibration signal time series;

[0102] S22, compressing the processed slurry pump vibration signal time series to the interval [-1, 1] to obtain a compressed slurry pump vibration signal time series A5, mapping the compression value in the compressed slurry pump vibration signal time series into an angle χ1, setting the time mapping radius to e1, and calculating the formula as follows:

[0103]

[0104] Where t represents the time corresponding to the processed slurry pump vibration signal time series, and δ represents the polar coordinate span coefficient;

[0105] The compressed values ​​in the compressed slurry pump vibration signal time series are again mapped into an angle χ2, and the Gram matrix C1=cos(χ1+χ2); in the Gram matrix, each data element represents the pixel value of a pixel point of the slurry pump vibration signal image, thereby generating a slurry pump vibration signal image;

[0106] S23, extracting texture features from the slurry pump vibration signal image using a Gauss-Markov random field, the specific steps are as follows:

[0107] S231, select pixel points on the slurry pump vibration signal image to obtain a pixel point set, set Represents the i4th pixel in the pixel set, represents the weight of the pixel point in the symmetric Gauss-Markov neighborhood, Represents Gaussian distribution noise, then the Gauss-Markov model calculation formula is as follows,

[0108]

[0109] Where D represents the Gauss-Markov model, g represents the number of pixel points, and i4 = 1, 2, 3, ..., g;

[0110] S232, substituting the pixel points of the slurry pump vibration signal image into the Gauss-Markov model to obtain a differential equation. The calculation formula is as follows:

[0111]

[0112] Where D′ represents the difference equation, represents the eigenvector, C2 represents the difference equation matrix;

[0113] Solving the difference equation error using a least square error method, selecting a 12-dimensional vector that satisfies the difference equation error, and recording the 12-dimensional vector that satisfies the difference equation error as a 12-dimensional eigenvector;

[0114] S3. Collect new slurry pump vibration signals to obtain new 12-dimensional feature vectors. Use the generator and discriminator to augment the new 12-dimensional feature vectors. Then, train the convolutional neural network and optimize the convolutional neural network parameters based on the electric eel foraging optimization algorithm to obtain a convolutional neural network recognition model to implement slurry pump fault detection.

[0115] The S3 includes the following steps:

[0116] S31. The deep convolutional generative adversarial network augmentation includes a generator and a discriminator. The generator consists of a reshape layer and a deconvolution layer. The activation function is set to the ReLU function. The data is input into the deep convolutional generative adversarial network through the reshape layer. The reshape layer reshapes the input data sample and then the deconvolution layer amplifies the feature to generate sample similarity features; the discriminator consists of a convolution layer and a fully connected layer. The input data sample and the sample similarity feature are subjected to feature extraction and dimensionality reduction processing. The sigmoid function is used as the activation function and the result is output after passing through the fully connected layer; the generator and the discriminator are alternately trained on the data samples until Nash equilibrium is reached to obtain augmented sample data;

[0117] A new slurry pump vibration signal is collected to obtain a new slurry pump vibration signal set. After outlier deletion and noise reduction processing are performed on the new slurry pump vibration signal set, a new slurry pump vibration signal image is generated to obtain a new 12-dimensional feature vector. The new 12-dimensional feature vector is divided into a feature training set and a feature test set. The feature training set is input into a generator to generate vibration signal similarity features. The vibration signal similarity features are then input into a discriminator to determine whether Nash equilibrium is reached. If Nash equilibrium is reached, the discriminator outputs an augmented feature training set. Otherwise, S31 is repeated until Nash equilibrium is reached.

[0118] S32, input the augmented feature training set into the convolutional neural network, and set the current number of iterations to The maximum number of iterations is G. When the current number of iterations reaches the maximum number of iterations, the iteration is stopped to obtain a trained convolutional neural network. The feature test set is input into the trained convolutional neural network, and the convolution kernel size and step size are optimized using the electric eel foraging optimization algorithm to obtain a convolutional neural network recognition model. The specific steps are as follows:

[0119] S321, using the trained convolutional neural network recognition accuracy as the fitness function, the electric eel population in the interaction stage, using stirring as the interaction sign in the search space, setting the current iteration number to h, the maximum iteration number to H, selecting a random electric eel j from the electric eel population, and setting the random electric eel h-th iteration position x j (h), the position of the i-th electric eel in the electric eel population at the h-th iteration The average position of the electric eels in the electric eel population at the hth iteration is recorded as ε represents the interaction parameter, then the position of the i-th electric eel in the electric eel population at the h+1th iteration is The calculation formula is as follows,

[0120] When the fitness function value of the random electric eel in the electric eel population is less than the fitness function value of the i-th electric eel in the electric eel population,

[0121] When the fitness function value of a random electric eel in the electric eel population is greater than or equal to the fitness function value of the i-th electric eel in the electric eel population,

[0122] When the electric eel population is in the resting stage, the position vector of the electric eel in the electric eel population is projected onto the main diagonal of the search space to establish a rest area; a three-dimensional coordinate axis is established for normalization, and the search space and the eel position are stipulated to be within the range of 0-1 on the three-dimensional coordinate axis. The projected position is regarded as the rest area, and the electric eel position corresponding to the best fitness function value of the hth iteration is set as The initial scale of the rest area is φ, and the scale of the rest area is The upper bound of the rest area is k′, the upper bound of the rest area is k″, the normalized number of the hth iteration is E(h), g1 represents a random number and g1∈(0,1), then the calculation formula for the h+1th iteration rest position x2(h+1) is as follows,

[0123]

[0124] After the resting position is determined, the electric eel population moves to the resting position. Let g2 represent a random number between 0 and 1 that obeys the normal distribution, g3 represent a random number and g3∈(0,1), round represents rounding, then the resting position of the i-th electric eel in the electric eel population at the h+1th iteration is The calculation formula is as follows,

[0125]

[0126] S322, in the hunting stage, the electric eels surround the prey and continuously accelerate the search for the local fitness function value. The electric eels first determine the pre-position of the hunting area and set the scale of the hunting area to γ1. The h-th iterative hunting position of the electric eels in the electric eel population is recorded as The prey position is x′, and the electric eel hunting range is The electric eel moves to the pre-position of the hunting area, and the pre-position of the hunting area in the hth iteration is set to x4(h). The calculation formula for the pre-position of the hunting area in the h+1th iteration is x5(h+1).

[0127]

[0128] After the pre-position of the hunting area is determined, the electric eel population moves to the pre-position of the hunting area. The curl factor is set to η, g4 represents a random number and g4∈(0,1), then the pre-position of the hunting area of ​​the i-th electric eel in the electric eel population in the h+1th iteration is The calculation formula is as follows,

[0129]

[0130] The electric eel population moves to the pre-position of the hunting area and begins the migration phase. At this time, the optimal fitness function value is continuously sought. The position of the h+1th iteration hunting area of ​​the i-th electric eel in the electric eel population is recorded as When the fitness function value of the i-th electric eel in the electric eel population at the h+1 iteration is less than the fitness function value of the i-th electric eel in the electric eel population at the h iteration, the position of the i-th electric eel in the electric eel population at the h+1 iteration is Otherwise, the position of the i-th electric eel in the electric eel population at the h+1th iteration is The electric eel continuously iterates its own position. When the current number of iterations is equal to the maximum number of iterations, the iteration stops and the final position of the i-th electric eel in the electric eel population is obtained. The coordinate values ​​of the two-dimensional coordinates of the final position of the i-th electric eel in the electric eel population correspond to the convolution kernel size and step size respectively;

[0131] S33. Input the 12-dimensional feature vector into a generator and a discriminator to obtain an augmented feature vector, and input the augmented feature vector into a convolutional neural network recognition model. If the convolutional neural network recognition model outputs 0, it indicates that the vibration signal of the slurry pump is abnormal and the slurry pump is faulty. If the convolutional neural network recognition model outputs 1, it indicates that the vibration signal of the slurry pump is normal and the slurry pump is not faulty, thereby completing the slurry pump fault detection.

[0132] S4. Obtain the original vibration signal of the slurry pump, obtain the IMF component of the original vibration signal of the slurry pump based on the empirical mode decomposition method, and train the LSTM neural network to obtain an LSTM neural network prediction model. Combined with the slurry pump vibration signal, the service life prediction value of the slurry pump is obtained to complete the slurry pump life prediction;

[0133] The S4 comprises the following steps:

[0134] S41, reacquire the original vibration signal of the slurry pump, set the upper envelope of the original vibration signal of the slurry pump to l1, the lower envelope of the original vibration signal of the slurry pump to l2, calculate the average value of the upper envelope of the original vibration signal of the slurry pump and the lower envelope of the original vibration signal of the slurry pump to obtain the average envelope of the vibration signal; use the original vibration signal of the slurry pump to subtract the average envelope of the vibration signal to obtain a new vibration signal component, calculate the average value of the upper envelope of the new vibration signal component and the lower envelope of the new vibration signal component value, recorded as a new vibration signal average envelope; when the new vibration signal average envelope is 0, and the absolute value of the difference between the number of zero points and the number of extreme value points of the new vibration signal average envelope on the time domain diagram is less than or equal to 1, the new vibration signal component is taken as the first IMF component; otherwise, S41 is repeated until the new vibration signal average envelope is 0, and the absolute value of the difference between the number of zero points and the number of extreme value points of the new vibration signal average envelope on the time domain diagram is less than or equal to 1, thereby obtaining the first IMF component;

[0135] S42, removing the first IMF component from the original vibration signal of the slurry pump, decomposing it to obtain a second original vibration signal of the slurry pump, performing S41 on the second original vibration signal of the slurry pump to obtain a second IMF component; continuously repeating S41 until the remaining original vibration signal of the slurry pump is a monotonic function, stopping the decomposition, recording the remaining original vibration signal of the slurry pump as an original residual vibration signal, and outputting the IMF component of the original vibration signal of the slurry pump and the original residual vibration signal;

[0136] In the LSTM neural network, the number of IMF components of the original vibration signal of the slurry pump is set to m, the sliding window size is n, the number of neurons is (m+2)×n, the current number of iterations is q, the maximum number of iterations is Q, the IMF components of the original vibration signal of the slurry pump, the original residual vibration signal and the original vibration signal of the slurry pump are used as the prediction sample set, the prediction sample set is divided into a prediction training set and a prediction test set, the prediction training set is input into the LSTM neural network, and when the current number of iterations reaches the maximum number of iterations, the trained LSTM neural network is obtained; the prediction test set is input into the trained LSTM neural network, and the error threshold is set to ψ. When the output error is less than ψ, the LSTM neural network prediction model is obtained, otherwise the weight is adjusted until the output error is less than ψ;

[0137] The IMF component of the slurry pump vibration signal is extracted to obtain the IMF component of the original vibration signal of the slurry pump and the residual vibration signal. The IMF component of the original vibration signal of the slurry pump, the residual vibration signal and the slurry pump vibration signal are used as a prediction verification set, and the prediction verification set is input into the LSTM neural network prediction model to output the predicted value of the slurry pump service life to complete the slurry pump life prediction.

[0138] This embodiment also discloses a system for slurry pump fault detection and life prediction based on big data, which specifically includes: a slurry pump vibration signal noise reduction module, a slurry pump vibration signal feature extraction module, a slurry pump fault detection module and a slurry pump life prediction module;

[0139] The slurry pump vibration signal noise reduction module is used to detect and delete abnormal vibration signals, and then achieve noise reduction after singular value decomposition and reconstruction signal processing;

[0140] The slurry pump vibration signal feature extraction module is used to generate a slurry pump vibration signal image based on the vibration signal, and obtain a 12-dimensional feature vector after feature extraction;

[0141] The slurry pump fault detection module is used to train a convolutional neural network to obtain a convolutional neural network recognition model to detect slurry pump faults;

[0142] The slurry pump life prediction module is used to train the LSTM neural network to obtain an LSTM neural network prediction model and output a predicted value of the slurry pump service life.

[0143] Throughout this specification, references to terms such as "one embodiment," "example," or "specific example" indicate that the specific features, structures, materials, or characteristics described in conjunction with that embodiment or example are included in at least one embodiment or example of the invention. In this specification, schematic representations of these terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.

[0144] The preferred embodiments of the invention disclosed above are intended only to help illustrate the invention. These preferred embodiments do not exhaust all details, nor do they limit the invention to the specific embodiments described. Obviously, many modifications and variations are possible based on the content of this specification. These embodiments are selected and described in detail in this specification to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention.

Claims

1. A method for slurry pump fault detection and life prediction based on big data, characterized in that: The steps include: S1. Collect the vibration signal of the slurry pump in the running state to obtain the slurry pump vibration signal, perform abnormal vibration signal detection on the slurry pump vibration signal based on the outlier factor algorithm, and delete the abnormal vibration signal; then perform singular value decomposition and reconstruction signal processing to complete the slurry pump vibration signal noise reduction and obtain the noise-reduced slurry pump vibration signal; S2, the slurry pump vibration signal after noise reduction is formed into a slurry pump vibration signal set, and then the segmented aggregation approximation and Gram angle field processing are performed to generate a slurry pump vibration signal image, and then feature extraction is performed based on the Gauss-Markov random field to obtain a 12-dimensional feature vector; S3. Collect new slurry pump vibration signals to obtain new 12-dimensional feature vectors. Use the generator and discriminator to augment the new 12-dimensional feature vectors. Then, train the convolutional neural network and optimize the convolutional neural network parameters to obtain a convolutional neural network recognition model to implement slurry pump fault detection. The S3 comprises the following steps: S31. Collect new slurry pump vibration signals to obtain a new slurry pump vibration signal set, obtain a new 12-dimensional feature vector, divide the new 12-dimensional feature vector into a feature training set and a feature test set, input the feature training set into a generator to generate vibration signal similarity features, and then input the vibration signal similarity features into a discriminator to determine whether Nash equilibrium is reached; if Nash equilibrium is reached, the discriminator outputs an augmented feature training set, otherwise S31 is repeated until Nash equilibrium is reached; S32, input the augmented feature training set into the convolutional neural network, and set the current number of iterations to The maximum number of iterations is G. When the current number of iterations reaches the maximum number of iterations, the iteration is stopped to obtain a trained convolutional neural network; the feature test set is input into the trained convolutional neural network, and the convolution kernel size and step size are optimized using the electric eel foraging optimization algorithm to obtain a convolutional neural network recognition model; S4. Obtain the original vibration signal of the slurry pump, obtain the IMF component of the original vibration signal of the slurry pump based on the empirical mode decomposition method, and train the LSTM neural network to obtain the LSTM neural network prediction model. Combined with the slurry pump vibration signal, the predicted value of the slurry pump service life is obtained to complete the slurry pump life prediction.

2. A method for slurry pump fault detection and life prediction based on big data according to claim 1, characterized in that: The S1 comprises the following steps: S11, when the slurry pump is in operation, collecting a vibration signal of the slurry pump to obtain a vibration signal of the slurry pump; S12. Calculate the outlier factor of the slurry pump vibration signal using an outlier factor algorithm, delete abnormal vibration signals according to the outlier factor, and obtain a set of abnormally deleted slurry pump vibration signals; S13. Performing singular value decomposition and signal reconstruction processing on the abnormally deleted slurry pump vibration signal set to obtain a noise-reduced slurry pump vibration signal.

3. The method for slurry pump fault detection and life prediction based on big data according to claim 2, characterized in that: The S2 comprises the following steps: S21, forming a noise-reduced slurry pump vibration signal set according to the noise-reduced slurry pump vibration signal, and obtaining a processed slurry pump vibration signal time series after performing segmented aggregation approximation processing; S22, generating a Gram matrix from the processed slurry pump vibration signal time series to obtain a slurry pump vibration signal image; S23. Perform texture feature extraction on the slurry pump vibration signal image using a Gauss-Markov random field to obtain a 12-dimensional feature vector.

4. The method for slurry pump fault detection and life prediction based on big data according to claim 3 is characterized in that: The S23 includes the following steps: S231, establishing a Gauss-Markov model on the slurry pump vibration signal image; S232. Using the differential equation of the Gauss-Markov model, select a 12-dimensional vector that satisfies the differential equation error and record it as a 12-dimensional feature vector.

5. The method for slurry pump fault detection and life prediction based on big data according to claim 4 is characterized in that: The 12-dimensional feature vector is input into the generator and the discriminator to obtain an augmented feature vector, which is then input into the convolutional neural network recognition model. If the convolutional neural network recognition model outputs 0, it indicates that the vibration signal of the slurry pump is abnormal and the slurry pump is faulty. If the convolutional neural network recognition model outputs 1, it indicates that the vibration signal of the slurry pump is normal and the slurry pump is not faulty, thus completing the slurry pump fault detection.

6. The method for slurry pump fault detection and life prediction based on big data according to claim 5, characterized in that: The S32 includes the following steps: S321, using the trained convolutional neural network recognition accuracy as the fitness function, the electric eel population in the interaction stage, using stirring as the interaction sign in the search space, setting the current iteration number to h, the maximum iteration number to H, selecting a random electric eel j from the electric eel population, and setting the random electric eel h-th iteration position x j (h), the position of the i-th electric eel in the electric eel population at the h-th iteration The average position of the electric eels in the electric eel population at the hth iteration is recorded as ε represents the interaction parameter, then the position of the i-th electric eel in the electric eel population at the h+1th iteration is The calculation formula is as follows, When the fitness function value of the random electric eel in the electric eel population is less than the fitness function value of the i-th electric eel in the electric eel population, When the fitness function value of a random electric eel in the electric eel population is greater than or equal to the fitness function value of the i-th electric eel in the electric eel population, When the electric eel population is in the resting stage, the position vector of the electric eel in the electric eel population is projected onto the main diagonal of the search space to establish a rest area. The resting position of the h+1th iteration is recorded as x2(h+1); After the resting position is determined, the electric eel population moves to the resting position, and the resting position of the i-th electric eel in the electric eel population at the h+1th iteration is recorded as S322, in the hunting phase of the electric eel population, the electric eels surround the prey and continuously accelerate the search for the local fitness function value. The pre-position of the hunting area in the h+1th iteration is recorded as x5(h+1); After the pre-position of the hunting area is determined, the electric eel population moves to the pre-position of the hunting area, and the pre-position of the hunting area of ​​the i-th electric eel in the electric eel population at the h+1th iteration is recorded as The electric eel population moves to the pre-position of the hunting area and begins the migration phase. At this time, the optimal fitness function value is continuously sought. The position of the h+1th iteration hunting area of ​​the i-th electric eel in the electric eel population is recorded as When the fitness function value of the i-th electric eel in the electric eel population at the h+1 iteration is less than the fitness function value of the i-th electric eel in the electric eel population at the h iteration, the position of the i-th electric eel in the electric eel population at the h+1 iteration is Otherwise, the position of the i-th electric eel in the electric eel population at the h+1th iteration is The electric eel continuously iterates its own position. When the current number of iterations is equal to the maximum number of iterations, the iteration is stopped to obtain the final position of the i-th electric eel in the electric eel population. The coordinate values ​​of the two-dimensional coordinates of the final position of the i-th electric eel in the electric eel population correspond to the convolution kernel size and step size respectively.

7. The method for slurry pump fault detection and life prediction based on big data according to claim 6, characterized in that: The S4 comprises the following steps: S41, reacquiring the original vibration signal of the slurry pump, extracting the IMF component of the original vibration signal of the slurry pump using the empirical mode decomposition method, and outputting the IMF component of the original vibration signal of the slurry pump and the original residual vibration signal; S42. In the LSTM neural network, the current number of iterations is set to q, the maximum number of iterations is set to Q, the IMF component of the original vibration signal of the slurry pump, the original residual vibration signal and the original vibration signal of the slurry pump are used as the prediction sample set, the prediction sample set is divided into a prediction training set and a prediction test set, the prediction training set is input into the LSTM neural network, and when the current number of iterations reaches the maximum number of iterations, the trained LSTM neural network is obtained; the prediction test set is input into the trained LSTM neural network, and the error threshold is set to ψ. When the output error is less than ψ, the LSTM neural network prediction model is obtained, otherwise the weight is adjusted until the output error is less than ψ.

8. The method for slurry pump fault detection and life prediction based on big data according to claim 7, characterized in that: The IMF components of the slurry pump vibration signal are extracted to obtain the IMF components of the original vibration signal of the slurry pump and the residual vibration signal. The IMF components of the original vibration signal of the slurry pump, the residual vibration signal and the slurry pump vibration signal are used as the prediction verification set, and the prediction verification set is input into the LSTM neural network prediction model to output the predicted value of the slurry pump service life to complete the slurry pump life prediction.

9. A system for implementing the method for slurry pump fault detection and life prediction based on big data according to any one of claims 1 to 8, characterized in that: Specifically include: Slurry pump vibration signal noise reduction module, slurry pump vibration signal feature extraction module, slurry pump fault detection module and slurry pump life prediction module; The slurry pump vibration signal noise reduction module is used to detect and delete abnormal vibration signals, and then achieve noise reduction after singular value decomposition and reconstruction signal processing; The slurry pump vibration signal feature extraction module is used to generate a slurry pump vibration signal image based on the vibration signal, and obtain a 12-dimensional feature vector after feature extraction; The slurry pump fault detection module is used to train a convolutional neural network to obtain a convolutional neural network recognition model to detect slurry pump faults; The slurry pump life prediction module is used to train the LSTM neural network to obtain an LSTM neural network prediction model and output a predicted value of the slurry pump service life.

Images