A novel intelligent fault diagnosis method for rotating machinery

Abstract

This invention provides a novel intelligent fault diagnosis method for rotating machinery, belonging to the field of fault diagnosis technology. The method includes: acquiring vibration signals of the rotating machinery under normal conditions and different fault types; preprocessing the vibration signals using Variational Mode Decomposition (VMD), decomposing and removing noise before reconstructing a new signal; converting the new signal into a two-dimensional time-frequency image using Short-Time Fourier Transform (STFT); optimizing the hyperparameters of a Convolutional Neural Network-Long Short-Time Memory (CNN-LSTM) using the Starfleet Optimization (SFA) algorithm; training the CNN-LSTM model by inputting the two-dimensional time-frequency image into the optimized hyperparameters; and inputting the fault diagnosis signal of the rotating machinery to be diagnosed, after noise reduction and two-dimensional time-frequency image conversion, into the VMD-SFA-CNN-LSTM model to achieve rotating machinery fault diagnosis. This invention has significant practical implications for improving the accuracy of rotating machinery fault diagnosis and guiding equipment maintenance.

Description

Technical Field

[0001] This invention relates to the field of rotating machinery fault diagnosis technology, and more specifically, to a novel intelligent fault diagnosis method for rotating machinery. Background Technology

[0002] Failures in rotating machinery can range from disrupting normal production and causing a decline in product quality to causing huge economic losses to enterprises, or even leading to major safety accidents and casualties. Therefore, timely and accurate diagnosis of rotating machinery failures is of great practical significance.

[0003] Currently, with the development of machine learning methods, artificial intelligence technology is being widely applied in various fields. In the field of rotating machinery fault diagnosis, applying machine learning methods, especially various neural network methods, has become one of the mainstream research directions. However, the performance of neural network models is affected by hyperparameters; inappropriate hyperparameters may directly lead to diagnostic errors and affect diagnostic accuracy. Therefore, optimizing neural networks using optimization algorithms to find the optimal matching of hyperparameters has become a current research hotspot. However, existing intelligent optimization algorithms still have room for improvement in terms of optimization accuracy and computational efficiency. Summary of the Invention

[0004] This invention provides a novel intelligent fault diagnosis method for rotating machinery. Fault diagnosis is based on VMD-SFA-CNN-LSTM, where Variational Mode Decomposition (VMD) is first used for signal denoising preprocessing, followed by the establishment of a Convolutional Neural Network-Long Short-Term Memory (CNN-LSTM) model optimized using the Star Fleet Algorithm (SFA). This invention innovatively proposes a novel Star Fleet algorithm (SFA) and uses it to optimize the hyperparameters of the CNN-LSTM, resulting in a lower iterative loss value and higher diagnostic accuracy, providing reliable technical support for improving the accuracy of rotating machinery fault diagnosis.

[0005] Humanity's fascination with space and interstellar travel has been long-standing, and numerous films and literary works have explored and imagined this concept. Different interstellar fleets from various organizations carry out various explorations and missions in the complex cosmic environment, with different types of interstellar ships within each fleet performing specific tasks to find higher-quality cosmic resources. The interstellar fleet optimization algorithm proposed in this invention is a metaheuristic swarm intelligence optimization algorithm created based on the inspiration drawn from the aforementioned fleet collaboration.

[0006] The technical solution to achieve the purpose of this invention is: a novel intelligent fault diagnosis method for rotating machinery, comprising the following steps:

[0007] Step 1: Acquire the vibration signals of the rotating machinery under normal conditions and different fault types for training;

[0008] Step 2: Variational mode decomposition is used to perform noise reduction preprocessing on the vibration signal. The preprocessing includes signal decomposition, noise reduction, and reconstruction to generate a new vibration signal.

[0009] Step 3: Convert the new vibration signal into a two-dimensional time-frequency image using a short-time Fourier transform;

[0010] Step 4: Set the parameters required for the Starfleet optimization algorithm;

[0011] Step 5: Using the two-dimensional time-frequency image as input, optimize the hyperparameters of the CNN-LSTM neural network through the Starfleet optimization algorithm;

[0012] Step 6: Based on the optimized hyperparameters, train a rotating machinery fault diagnosis model based on a novel intelligent fault diagnosis method for rotating machinery using VMD-SFA-CNN-LSTM.

[0013] Step 7: Denoise the fault diagnosis signal to be diagnosed, convert it into a two-dimensional time-frequency image through short-time Fourier transform, and input the image into the trained VMD-SFA-CNN-LSTM intelligent rotating machinery fault diagnosis model to realize rotating machinery fault diagnosis.

[0014] Furthermore, the novel intelligent diagnostic method for rotating machinery faults is characterized in that, in step four, the parameters required by the interstellar fleet optimization algorithm include the total number of ships. M Maximum number of iterations tmax Number of iterations during the initial resource exploration phase n Periodic fleet reconfiguration strategy cycle m ;

[0015] Furthermore, the novel intelligent diagnostic method for rotating machinery faults is characterized in that, in step four, the interstellar fleet optimization algorithm uses three types of ships to represent the optimization individual: a space-breaking ship, a dispatch ship, and a search ship, and the optimization is divided into four stages:

[0016] (1) Initialization phase: Random generation using chaotic mapping. M Several ships, and the initialization phase. M All search ships are classified as air-piercing ships, and their expression is:

[0017] x i,0 =rand()∈[0,1]

[0018] x i,k+1 =x i,k / 0.7 when x i,k <0.7

[0019] x i,k+1 = (1- x i,k ) / 0.7 when x i,k ≥0.7

[0020] X i =lb+B i ⊙(ub-lb)

[0021] in, x i,0 Representative ships i The initial value of the corresponding chaotic sequence, rand()∈[0,1], represents taking a random number between 0 and 1; x i,k and x i,k+1 Representative ships i corresponding dim In the dimensional chaotic sequence, the first k peacekeeping k +1 dimension value, k From 0 to dim Integer value of -1, i =1,2,..., M X i Representing the i The position of the ship, B i =( x i,1 , x i,2 ,..., x i,dim (for ships) i The corresponding chaotic sequence, ⊙ represents element-wise multiplication, lb is the lower bound of the search domain, ub is the upper bound of the search domain, and X i ,lb,ub are all dim A dimensional vector, where dim Represents the dimension of the problem to be solved;

[0022] (2) Initial resource exploration stage: The total number of iterations in this stage is n And in each iteration, each air-breaking ship will select N Update the position of each dimension. N The expression is:

[0023] N= max(1,ceil( dim / 4))

[0024] Among them, the N The position of each dimension is determined by all of the spaceship's... dim Randomly select 2 from each dimension N The values ​​of each dimension are combined in pairs; ceil() is an integer parenthesis, which means that the contents of the parentheses are rounded up; max() means that the maximum value of all the values ​​in the parentheses is taken.

[0025] When facing ships i When updating the position based on the dimension, another ship will be randomly selected as the reference ship. r The update method is as follows:

[0026] Δ j =X r (j) -X i (j)

[0027] Δ k =X r (k) -X i (k)

[0028] X i new(j) =X i (j) + or 1 s j Δ j

[0029] X i new(k) =X i (k) + or 2 s k Δ k

[0030] in, j and k This represents the two dimensions of the value to be updated; X r (j) X r (k) For reference ship r In dimensions j and k Location, X i (j) X i (k) Ships awaiting update i In dimensions j and k Position, Δj Δ k For reference ship r and ships i The difference between the two dimensions, or 1, or 2 A random number between 0 and 1 s j , s k ∈{-1,1} is a random direction factor, representing a random value equal to -1 or 1, X i new(j) X i new(k) For ships i In dimensions j and k Updated location;

[0031] (3) Periodic Fleet Reconstruction Phase: The first iteration after the initial resource exploration phase immediately triggers periodic fleet reconstruction, and this reconstruction continues thereafter. m Each iteration triggers a periodic fleet reconfiguration, which is performed as follows:

[0032] First, a resource assessment is conducted on all ships, using their individual historical best fitness values ​​as the assessment criterion, dividing them into leading ships, stern ships, and midship ships:

[0033] H =max(2,floor(0.1 M ))

[0034] T =max(2,floor(0.1 M ))

[0035] M mid = M - H - T

[0036] Here, floor() represents the floor function, which rounds down the contents of the parentheses to the nearest integer. H Number of leading ships T The number of ships at the stern. M mid The number of ships in the central region;

[0037] Secondly, the ship types are reconstructed according to their location:

[0038] ① For the leading ship, it transforms into a dispatch ship and jumps to its own optimal position, and assigns search ship attraction weights to it. The weights include the weights currently assigned by the dispatch ship.w i Normalized probability of selection p i Cumulative probability distribution P i ， as follows:

[0039] X i new =P i

[0040] r =1+0.5( t / tmax )

[0041] w i =( H - i ) r when I <H / 3

[0042] w i = H - i when i≥H / 3

[0043] p i = w i / sum( w 1, w 2,... w H ) i =1,2,…, H

[0044] P i = sum( p 1, p 2,..., p i ) i =1,2,…, H

[0045] Among them, P i For the first i The best historical position of an individual ship t This represents the current iteration number. tmax The maximum number of iterations, r As a ranking enhancement factor, sum() represents summing all the values ​​within the parentheses;

[0046] ② For ships at the stern, the transformation into a space-breaking ship and the jump to a random location in space is expressed as follows:

[0047] X i new =lb+r⊙(ub-lb)

[0048] Among them, r~U dim (0,1) represents each dimension consisting of random numbers from 0 to 1. dim dimensional vector;

[0049] ③ For mid-range ships, based on the probability factor of mid-range ship type. α It can be converted into a skybreaker or a search ship, with the probability of being converted into a skybreaker being:

[0050] α =0.6-0.5min(1, t / 0.75 tmax )

[0051] Here, min() represents taking the minimum value among all values ​​within the parentheses, so 1- α The probability of converting into a search ship;

[0052] When a ship in the middle changes its type to a search vessel, the cumulative probability distribution of the dispatching vessel is used. P i To select a dispatch ship for it, the expression is:

[0053] X selected =X b

[0054] in, b =min{ i= 1,2,…, H | P i >= or 3}, min{ i= 1,2,…, H | P i >= or 3} represents all that meet the criteria P i >= or 3 of i Take the minimum value among the values, X b For the selected dispatch ship, or 3 A random number between 0 and 1;

[0055] (4) Fleet Coordination Phase: Requires three ship types to explore together:

[0056] ① The air-breaking ship will be updated in the same way as the initial resource exploration phase;

[0057] ② The dispatching ship will randomly select Ns The set of dimensions to be updated consists of [number] dimensions. Q Perform location update. Ns and Q The expression is:

[0058] Ns =max(1,⌈ dim / 2⌉)

[0059] Q ={ j 1,..., j Ns}

[0060] For each dimension to be updated j ∈ Q Its update formula is:

[0061] C =0.1+0.9 t / tmax

[0062] X i new(j) =X i (j) + C (G (j) -X i (j) )+ or 4 r 1(1- C (X) i (j) -X p (j) )+ or 5(1- r 1)(1- C (X) q (j) -X i (j) )

[0063] in, C For the adaptive parameters of the scheduling ship, r 1∈{0,1} is a binary random selection factor. p For randomly selected non-self-managed ship indexes, q Let G be the index of any randomly selected ship, and G be the globally optimal ship position. or 4, or 5 is a random number between 0 and 1;

[0064] ③ The search ship's movement modes are divided into jump acceleration mode and a hybrid curve and jump mode:

[0065] when r When 2 < 0.5, a jump acceleration mode is adopted, and the update method is as follows:

[0066] D=[ d 1,..., d dim ], d j ∈{-1, 1}

[0067] Δ norm =||(F i -X i ) / (ub-lb)||⊙(ub-lb)

[0068] X i base =F i + or 6D⊙Δ norm

[0069] X i new =X i base +(1- t / tmax ) 1 / 2 or 7(F i -X i base )+( t / tmax ) 1 / 2 or 8(GX i base )

[0070] in, r 2∈(0,1) represents the engine mode parameter, and D is a combination of -1 and 1. dim dimensional direction vector, F i For the first i The optimal position of the fleet to which the ships belong, Δ norm For the first i The distance vector of each ship from the optimal position of the fleet in each dimension, X i base X represents a jump position in this engine mode. i new For search ship i Updated location; or 6, or 7, or 8 is a random number between 0 and 1;

[0071] when r When 2>=0.5, a mixed mode of curve and jump is adopted, and when L < t / tmax When using curve mode, the expression is as follows:

[0072] X i new =X i + or 9(F i -X i )+ or 10 (GX i -0.01 or 11 (X u -X i )

[0073] in, L , or 9, or 10 , or 11 All are random numbers between 0 and 1. u For non-first i The random dispatcher index of the fleet dispatcher for each ship.

[0074] when L > t / tmax When using jump mode, it is the same as one jump in the jump acceleration mode section:

[0075] D=[ d 1,..., d dim ], d j ∈{-1, 1}

[0076] Δ norm =||(F i -X i ) / (ub-lb)||⊙(ub-lb)

[0077] X i new =F i + or 12 D⊙Δ norm

[0078] in, or 12 A random number between 0 and 1.

[0079] Furthermore, the novel intelligent diagnostic method for rotating machinery faults is characterized in that, in step five, the CNN-LSTM neural network includes four convolutional layers, three pooling layers, one flattening layer, one LSTM layer, and one output layer. The CNN model shape is such that each of the first three convolutional layers is followed by a pooling layer, and the last convolutional layer is followed by a flattening layer. The data is then input into the LSTM model and then into the output layer. The activation function of the convolutional layers is the ReLU function, with a stride of (2, 2) and zero padding. The pooling layer size is 2×2. The output layer outputs 4, and the activation function is the softmax function. The hyperparameters include the learning rate and the L2 regularization parameter.

[0080] The beneficial effects of this invention are as follows: This invention proposes a novel metaheuristic swarm intelligence optimization algorithm—the Starfleet Optimization Algorithm (SFA). Its innovation lies in effectively balancing global exploration and local development by simulating the collaborative working mechanism of three types of heterogeneous intelligent agents (air-breaking ships, scheduling ships, and search ships). The algorithm employs a dimensionality-pair differential evolution strategy to enhance solution space exploration capabilities, combines adaptive parameter-guided multi-source information fusion to improve regional development efficiency, and pioneers a periodic fleet reconstruction mechanism. During scheduling, it achieves intelligent adjustment of the optimization process through elite individual guidance, inefficient individual reset, and dynamic type conversion strategies. Experimental verification shows that this method exhibits significant improvements in convergence speed, solution quality, and stability. This invention optimizes the hyperparameters of CNN-LSTM based on the SFA algorithm, enabling the optimized CNN-LSTM to extract fault features more effectively and improving the accuracy of rotating machinery fault diagnosis. Attached Figure Description

[0081] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained from these drawings without creative effort.

[0082] Figure 1 This is a flowchart of a novel intelligent fault diagnosis method for rotating machinery.

[0083] Figure 2 This is a diagram of the CNN-LSTM model structure;

[0084] Figure 3 This is a detailed flowchart of the SFA optimization algorithm;

[0085] Figure 4 This is a confusion matrix diagram of the fault diagnosis results of the VMD-CNN-LSTM method;

[0086] Figure 5This is a confusion matrix diagram of the fault diagnosis results of the VMD-PSO-CNN-LSTM method;

[0087] Figure 6 This is a confusion matrix diagram of the fault diagnosis results of the VMD-SFA-CNN-LSTM method;

[0088] Figure 7 This is a comparison chart of convergence in PSO-SFA parameter optimization;

[0089] Figure 8 This is a comparison chart of the accuracy rates of different methods. Detailed Implementation

[0090] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the implementation examples and accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to represent selected embodiments of the invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0091] The method of the present invention will be illustrated below using a specific case of rotating machinery fault diagnosis as an example.

[0092] like Figure 1 A novel intelligent fault diagnosis method for rotating machinery includes the following steps:

[0093] Step 1: Acquire the vibration signals of the rotating machinery under normal conditions and different fault types for training;

[0094] Step 2: Variational mode decomposition is used to perform noise reduction preprocessing on the vibration signal. The preprocessing includes signal decomposition, noise reduction, and reconstruction to generate a new vibration signal.

[0095] Step 3: Convert the new vibration signal into a two-dimensional time-frequency image using a short-time Fourier transform;

[0096] Step 4: Set the parameters required for the Starfleet optimization algorithm;

[0097] Step 5: Using the two-dimensional time-frequency image as input, optimize the hyperparameters of the CNN-LSTM neural network through the Starfleet optimization algorithm;

[0098] Step 6: Based on the optimized hyperparameters, train a rotating machinery fault diagnosis model based on a novel intelligent fault diagnosis method for rotating machinery using VMD-SFA-CNN-LSTM.

[0099] Step 7: Denoise the fault diagnosis signal to be diagnosed, convert it into a two-dimensional time-frequency image through short-time Fourier transform, and input the image into the trained VMD-SFA-CNN-LSTM intelligent rotating machinery fault diagnosis model to realize rotating machinery fault diagnosis.

[0100] Furthermore, in the novel intelligent diagnostic method for rotating machinery faults, step four requires parameters for the interstellar fleet optimization algorithm, including the total number of ships. M Maximum number of iterations tmax Number of iterations during the initial resource exploration phase n Periodic fleet reconfiguration strategy cycle m ;

[0101] Furthermore, in the novel intelligent diagnostic method for rotating machinery faults, in step four, the interstellar fleet optimization algorithm uses three types of ships to represent the optimization entity: a space-breaking ship, a dispatching ship, and a search ship. The optimization process is divided into four stages:

[0102] (1) Initialization phase: Random generation using chaotic mapping. M Several ships, and the initialization phase. M All search ships are classified as air-piercing ships, and their expression is:

[0103] x i,0 =rand()∈[0,1]

[0104] x i,k+1 = x i,k / 0.7 when x i,k <0.7

[0105] x i,k+1 = (1- x i,k ) / 0.7 when x i,k ≥0.7

[0106] X i =lb+B i ⊙(ub-lb)

[0107] in, x i,0 Representative ships iThe initial value of the corresponding chaotic sequence, rand()∈[0,1], represents taking a random number between 0 and 1; x i,k and x i,k+1 Representative ships i corresponding dim In the dimensional chaotic sequence, the first k peacekeeping k +1 dimension value, k From 0 to dim Integer value of -1, i =1,2,..., M X i Representing the i The position of the ship, B i =( x i,1 , x i,2 ,..., x i,dim (for ships) i The corresponding chaotic sequence, ⊙ represents element-wise multiplication, lb is the lower bound of the search domain, ub is the upper bound of the search domain, and X i ,lb,ub are all dim A dimensional vector, where dim Represents the dimension of the problem to be solved;

[0108] (2) Initial resource exploration stage: The total number of iterations in this stage is n And in each iteration, each air-breaking ship will select N Update the position of each dimension. N The expression is:

[0109] N= max(1,ceil( dim / 4))

[0110] Among them, the N The position of each dimension is determined by all of the spaceship's... dim Randomly select 2 from each dimension N The values ​​of each dimension are combined in pairs; ceil() is an integer parenthesis, which means that the contents of the parentheses are rounded up; max() means that the maximum value of all the values ​​in the parentheses is taken.

[0111] When facing ships i When updating the position based on the dimension, another ship will be randomly selected as the reference ship. r The update method is as follows:

[0112] Δ j =X r (j) -Xi (j)

[0113] Δ k =X r (k) -X i (k)

[0114] X i new(j) =X i (j) + or 1 s j Δ j

[0115] X i new(k) =X i (k) + or 2 s k Δ k

[0116] in, j and k This represents the two dimensions of the value to be updated; X r (j) X r (k) For reference ship r In dimensions j and k Location, X i (j) X i (k) Ships awaiting update i In dimensions j and k Position, Δ j Δ k For reference ship r and ships i The difference between the two dimensions, or 1, or 2 A random number between 0 and 1 s j , s k ∈{−1,1} is a random direction factor, representing a random value equal to -1 or 1, X i new(j) X i new(k) For ships i In dimensions j and k Updated location;

[0117] (3) Periodic Fleet Reconstruction Phase: The first iteration after the initial resource exploration phase immediately triggers periodic fleet reconstruction, and this reconstruction continues thereafter. m Each iteration triggers a periodic fleet reconfiguration, which is performed as follows:

[0118] First, a resource assessment is conducted on all ships, using their individual historical best fitness values ​​as the assessment criterion, dividing them into leading ships, stern ships, and midship ships:

[0119] H =max(2,floor(0.1 M ))

[0120] T =max(2,floor(0.1 M ))

[0121] M mid = M - H - T

[0122] Here, floor() represents the floor function, which rounds down the contents of the parentheses to the nearest integer. H Number of leading ships T The number of ships at the stern. M mid The number of ships in the central region;

[0123] Secondly, the ship types are reconstructed according to their location:

[0124] ① For the leading ship, it transforms into a dispatch ship and jumps to its own optimal position, and assigns search ship attraction weights to it. The weights include the weights currently assigned by the dispatch ship. w i Normalized probability of selection p i Cumulative probability distribution P i ， as follows:

[0125] X i new =P i

[0126] r =1+0.5( t / tmax )

[0127] w i =( H -i ) r when I <H / 3

[0128] w i = H - i when i≥H / 3

[0129] p i = w i / sum( w 1, w 2,... w H ) i =1,2,…, H

[0130] P i = sum( p 1, p 2,..., p i ) i =1,2,…, H

[0131] Among them, P i For the first i The best historical position of an individual ship t This represents the current iteration number. tmax The maximum number of iterations, r As a ranking enhancement factor, sum() represents summing all the values ​​within the parentheses;

[0132] ② For ships at the stern, the transformation into a space-breaking ship and the jump to a random location in space is expressed as follows:

[0133] X i new =lb+r⊙(ub-lb)

[0134] Among them, r~U dim (0,1) represents each dimension consisting of random numbers from 0 to 1. dim dimensional vector;

[0135] ③ For mid-range ships, based on the probability factor of mid-range ship type. α It can be converted into a skybreaker or a search ship, with the probability of being converted into a skybreaker being:

[0136] α =0.6-0.5min(1, t / 0.75 tmax )

[0137] Here, min() represents taking the minimum value among all values ​​within the parentheses, so 1- α The probability of converting into a search ship;

[0138] When a ship in the middle changes its type to a search vessel, the cumulative probability distribution of the dispatching vessel is used. P i To select a dispatch ship for it, the expression is:

[0139] X selected =X b

[0140] in, b =min{ i= 1,2,…, H | P i >= or 3}, min{ i= 1,2,…, H | P i >= or 3} represents all that meet the criteria P i >= or 3 of i Take the minimum value among the values, X b For the selected dispatch ship, or 3 A random number between 0 and 1;

[0141] (4) Fleet Coordination Phase: Requires three ship types to explore together:

[0142] ① The air-breaking ship will be updated in the same way as the initial resource exploration phase;

[0143] ② The dispatching ship will randomly select Ns The set of dimensions to be updated consists of [number] dimensions. Q Perform location update. Ns and Q The expression is:

[0144] Ns =max(1,⌈ dim / 2⌉)

[0145] Q ={ j 1,..., j Ns}

[0146] For each dimension to be updatedj ∈ Q Its update formula is:

[0147] C =0.1+0.9 t / tmax

[0148] X i new(j) =X i (j) + C (G (j) -X i (j) )+ or 4 r 1(1- C (X) i (j) -X p (j) )+ or 5(1- r 1)(1- C (X) q (j) -X i (j) )

[0149] in, C For the adaptive parameters of the scheduling ship, r 1∈{0,1} is a binary random selection factor. p For randomly selected non-self-managed ship indexes, q Let G be the index of any randomly selected ship, and G be the globally optimal ship position. or 4, or 5 is a random number between 0 and 1;

[0150] ③ The search ship's movement modes are divided into jump acceleration mode and a hybrid curve and jump mode:

[0151] when r When 2 < 0.5, a jump acceleration mode is adopted, and the update method is as follows:

[0152] D=[ d 1,..., d dim ], d j ∈{-1, 1}

[0153] Δ norm =||(F i -X i ) / (ub-lb)||⊙(ub-lb)

[0154] X ibase =F i + or 6D⊙Δ norm

[0155] X i new =X i base +(1- t / tmax ) 1 / 2 or 7(F i -X i base )+( t / tmax ) 1 / 2 or 8(GX i base )

[0156] in, r 2∈(0,1) represents the engine mode parameter, and D is a combination of -1 and 1. dim dimensional direction vector, F i For the first i The optimal position of the fleet to which the ships belong, Δ norm For the first i The distance vector of each ship from the optimal position of the fleet in each dimension, X i base X represents a jump position in this engine mode. i new For search ship i Updated location; or 6, or 7, or 8 is a random number between 0 and 1;

[0157] when r When 2>=0.5, a mixed mode of curve and jump is adopted, and when L < t / tmax When using curve mode, the expression is as follows:

[0158] X i new =X i + or 9(F i -X i )+ or 10 (GX i -0.01 or 11 (X u -X i )

[0159] in, L , or 9, or 10 , or 11 All are random numbers between 0 and 1. u For non-first i The random dispatcher index of the fleet dispatcher for each ship.

[0160] when L > t / tmax When using jump mode, it is the same as one jump in the jump acceleration mode section:

[0161] D=[ d 1,..., d dim ], d j ∈{-1, 1}

[0162] Δ norm =||(F i -X i ) / (ub-lb)||⊙(ub-lb)

[0163] X i new =F i + or 12 D⊙Δ norm

[0164] in, or 12 A random number between 0 and 1.

[0165] Furthermore, the novel intelligent fault diagnosis method for rotating machinery is characterized in that, in step five, the CNN-LSTM neural network, as shown... Figure 2 As shown, the model includes 4 convolutional layers, 3 pooling layers, 1 flattening layer, 1 LSTM layer, and 1 output layer. The shape of the CNN model is such that each of the first three convolutional layers is followed by a pooling layer, and the last convolutional layer is followed by a flattening layer. The input is then fed into the LSTM model and then into the output layer. The activation function of the convolutional layers is the ReLU function, the stride is (2, 2), and zero padding is used. The pooling layer size is 2×2. The output layer outputs 4, and the activation function is the softmax function. The hyperparameters include the learning rate and the L2 regularization parameter.

[0166] The experimental data comes from the XJTU-SY rolling bearing accelerated life test, and -12dB Gaussian noise was added to the input vibration signal. The experimental data information is shown in Table 1.

[0167]

[0168] The specific implementation process is as follows:

[0169] A novel intelligent fault diagnosis method for rotating machinery includes the following steps:

[0170] Step 1: Obtain the vibration signals of the rotating machinery under normal conditions and different fault types. In this embodiment, 20 samples of each fault state are randomly selected, accounting for 70% of the total 80 samples as training samples, 30% as verification samples, and 30% of the total 400 samples as test samples.

[0171] Step 2: Variational mode decomposition is used to perform noise reduction preprocessing on the vibration signal. The preprocessing includes signal decomposition, noise reduction, and reconstruction to generate a new vibration signal.

[0172] Step 3: Convert the new vibration signal into a two-dimensional time-frequency image using a short-time Fourier transform;

[0173] Step 4: Set the parameters required for the Starfleet Optimization Algorithm (SFA), including the total number of ships. M =20, maximum number of iterations tmax =80, n =8, m =8;

[0174] Step 5: Using the two-dimensional time-frequency image as input, optimize the hyperparameters of the CNN-LSTM neural network through the Starfleet optimization algorithm. The CNN-LSTM model structure is as follows: Figure 2 As shown; the specific process of the SFA algorithm is as follows: Figure 3 As shown;

[0175] Step 6: Based on the optimized hyperparameters, train a rotating machinery fault diagnosis model based on a novel intelligent fault diagnosis method for rotating machinery using VMD-SFA-CNN-LSTM.

[0176] Step 7: Denoise the fault diagnosis signal to be diagnosed, convert it into a two-dimensional time-frequency image through short-time Fourier transform, and input the image into the trained VMD-SFA-CNN-LSTM intelligent rotating machinery fault diagnosis model to realize rotating machinery fault diagnosis.

[0177] To verify the advantages of the method of this invention, after VMD noise reduction preprocessing and STFT conversion of two-dimensional time-frequency images, calculations were performed using a conventional convolutional neural network-long short-term memory network (CNN-LSTM) and a CNN-LSTM optimized by particle swarm optimization (PSO), respectively, and the results were compared with the method of this invention. The accuracy and F1 score of each method were used to evaluate the classification results, and the evaluation results are shown in Table 2. Figure 4 , Figure 5 , Figure 6 , Figure 7 , Figure 8 It is evident that the VMD-SFA-CNN-LSTM method proposed in this invention outperforms the compared algorithms in terms of accuracy, F1 score, and convergence of optimization loss. Figure 7 This is a comparison chart of the convergence of different methods for optimizing loss.

[0178]

[0179] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the invention by those skilled in the art. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the invention should be included within the scope of protection of the invention.

Claims

1. A novel intelligent fault diagnosis method for rotating machinery, characterized in that, Includes the following steps: Step 1: Acquire the vibration signals of the rotating machinery under normal conditions and different fault types for training; Step 2: Variational mode decomposition is used to perform noise reduction preprocessing on the vibration signal. The preprocessing includes signal decomposition, noise reduction, and reconstruction to generate a new vibration signal. Step 3: Convert the new vibration signal into a two-dimensional time-frequency image using a short-time Fourier transform; Step 4: Set the parameters required for the Starfleet optimization algorithm; Step 5: Using the two-dimensional time-frequency image as input, optimize the hyperparameters of the CNN-LSTM neural network through the Starfleet optimization algorithm; Step 6: Based on the optimized hyperparameters, train a rotating machinery fault diagnosis model based on a novel intelligent fault diagnosis method for rotating machinery using VMD-SFA-CNN-LSTM. Step 7: Denoise the fault diagnosis signal to be diagnosed, convert it into a two-dimensional time-frequency image through short-time Fourier transform, and input the image into the trained VMD-SFA-CNN-LSTM intelligent rotating machinery fault diagnosis model to realize rotating machinery fault diagnosis. The Starfleet optimization algorithm uses three types of ships to represent the optimization targets: space-breaking ships, dispatch ships, and search ships. The optimization process consists of four stages: (1) Initialization phase: Random generation using chaotic mapping. M Several ships, and the initialization phase. M All search ships are classified as air-piercing ships, and their expression is: x i,0 =rand()∈[0,1] x i,k+1 = x i,k / 0.7 when x i,k <0.7 x i,k+1 = (1- x i,k ) / 0.7 when x i,k ≥0.7 X i =lb+B i ⊙(ub-lb) in, x i,0 Representative ships i The initial value of the corresponding chaotic sequence, rand()∈[0,1], represents taking a random number between 0 and 1; x i,k and x i,k+1 Representative ships i corresponding dim In the dimensional chaotic sequence, the first k peacekeeping k +1 dimension value, k From 0 to dim Integer value of -1, i =1,2,..., M X i Representing the i The position of the ship, B i =( x i,1 , x i,2 ,..., x i,dim (for ships) i The corresponding chaotic sequence, ⊙ represents element-wise multiplication, lb is the lower bound of the search domain, ub is the upper bound of the search domain, and X i ,lb,ub are all dim A dimensional vector, where dim Represents the dimension of the problem to be solved; (2) Initial resource exploration stage: The total number of iterations in this stage is n And in each iteration, each air-breaking ship will select N Update the position of each dimension. N The expression is: N= max(1,ceil( dim / 4)) Among them, the N The position of each dimension is determined by all of the spaceship's... dim Randomly select 2 from each dimension N The values ​​of each dimension are combined in pairs; ceil() represents rounding up the contents of the parentheses; max() represents taking the maximum value of all the values ​​in the parentheses; When facing ships i When updating the position based on the dimension, another ship will be randomly selected as the reference ship. r The update method is as follows: Δ j =X r (j) -X i (j) Δ k =X r (k) -X i (k) X i new(j) =X i (j) + η 1 σ j Δ j X i new(k) =X i (k) + η 2 σ k Δ k in, j and k This represents the two dimensions of the value to be updated; X r (j) X r (k) For reference ship r In dimensions j and k Location, X i (j) X i (k) Ships awaiting update i In dimensions j and k Position, Δ j Δ k For reference ship r and ships i The difference between the two dimensions, η 1, η 2 A random number between 0 and 1 σ j , σ k ∈{−1,1} is a random direction factor, representing a random value equal to -1 or 1, X i new (j) X i new(k) For ships i In dimensions j and k Updated location; (3) Periodic Fleet Reconstruction Phase: The first iteration after the initial resource exploration phase immediately triggers periodic fleet reconstruction, and this reconstruction continues thereafter. m Each iteration triggers a periodic fleet reconfiguration, which is performed as follows: First, a resource assessment is conducted on all ships, using their individual historical best fitness values ​​as the assessment criterion, dividing them into leading ships, stern ships, and midship ships: H =max(2,floor(0.1 M )) T =max(2,floor(0.1 M )) M mid = M - H - T Here, floor() represents rounding down the content within the parentheses. H Number of leading ships T The number of ships at the stern. M mid The number of ships in the central region; Secondly, the ship types are reconstructed according to their location: ① For the leading ship, it transforms into a dispatch ship and jumps to its own optimal position, and assigns search ship attraction weights to it. The weights include the weights currently assigned by the dispatch ship. w i Normalized probability of selection p i Cumulative probability distribution P i ， as follows: X i new =P i ρ =1+0.5( t / tmax ) w i =( H - i ) ρ when i <H / 3 w i = H - i when i≥H / 3 p i = w i / sum( w 1, w 2,... w H ) i =1,2,…, H P i = sum( p 1, p 2,..., p i ) i =1,2,…, H Among them, P i For the first i The best historical position of an individual ship t This represents the current iteration number. tmax The maximum number of iterations, ρ As a ranking enhancement factor, sum() represents summing all the values ​​within the parentheses; ② For ships at the stern, the transformation into a space-breaking ship and the jump to a random location in space is expressed as follows: X i new =lb+r⊙(ub-lb) Among them, r~U dim (0,1) represents each dimension consisting of random numbers from 0 to 1. dim dimensional vector; ③ For mid-range ships, based on the probability factor of mid-range ship type. α It can be converted into a skybreaker or a search ship, with the probability of being converted into a skybreaker being: α =0.6-0.5min(1, t / 0.75 tmax ) Here, min() represents taking the minimum value among all values ​​within the parentheses, so 1- α The probability of converting into a search ship; When a ship in the middle changes its type to a search vessel, the cumulative probability distribution of the dispatching vessel is used. P i To select a dispatch ship for it, the expression is: X selected =X b in, b =min{ i= 1,2,…, H | P i >= η 3 }, min{ i= 1,2,…, H | P i >= η 3 } represents all that meet the criteria P i >= η 3 of i Take the minimum value among the values, X b For the selected dispatch ship, η 3 A random number between 0 and 1; (4) Fleet Coordination Phase: Requires three ship types to explore together: ① The air-breaking ship will be updated in the same way as the initial resource exploration phase; ② The dispatching ship will randomly select Ns The set of dimensions to be updated consists of [number] dimensions. Q Perform location update. Ns and Q The expression is: Ns =max(1,ceil( dim / 2)) Q ={ j 1,..., j Ns} For each dimension to be updated j ∈ Q Its update formula is: C =0.1+0.9 t / tmax X i new(j) =X i (j) + C (G (j) -X i (j) )+ η 4 r 1(1- C )(X i (j) -X p (j) )+ η 5(1- r 1)(1- C )(X q (j) -X i (j) ) in, C For the adaptive parameters of the scheduling ship, r 1∈{0,1} is a binary random selection factor. p For randomly selected non-self-managed ship indexes, q Let G be the index of any randomly selected ship, and G be the globally optimal ship position. η 4, η 5 is a random number between 0 and 1; ③ The search ship's movement modes are divided into jump acceleration mode and a hybrid curve and jump mode: when r When 2 < 0.5, a jump acceleration mode is adopted, and the update method is as follows: D=[ d 1,..., d dim ]， d j ∈{-1，1} Δ norm =||(F i -X i ) / (ub-lb)||⊙(ub-lb) X i base =F i + η 6D⊙Δ norm X i new =X i base +(1- t / tmax ) 1 / 2 η 7(F i -X i base )+( t / tmax ) 1 / 2 η 8(G-X i base ) in, r 2∈(0,1) represents the engine mode parameter, and D is a combination of -1 and 1. dim dimensional direction vector, F i For the first i The optimal position of the fleet to which the ships belong, Δ norm For the first i The distance vector of each ship from the optimal position of the fleet in each dimension, X i base X represents a jump position in this engine mode. i new For search ship i Updated location; η 6, η 7, η 8 is a random number between 0 and 1; when r When 2>=0.5, a mixed mode of curve and jump is adopted, and when L < t / tmax When using curve mode, the expression is as follows: X i new =X i + η 9(F i -X i )+ η 10 (G-X i )-0.01 η 11 (X u -X i ) in, L , η 9, η 10 , η 11 All are random numbers between 0 and 1. u For non-first i The random dispatcher index of the fleet dispatcher for each ship. when L > t / tmax When using jump mode, it is the same as one jump in the jump acceleration mode section: D=[ d 1,..., d dim ]， d j ∈{-1，1} Δ norm =||(F i -X i ) / (ub-lb)||⊙(ub-lb) X i new =F i + η 12 D⊙Δ norm in, η 12 A random number between 0 and 1.

2. The novel intelligent fault diagnosis method for rotating machinery as described in claim 1, characterized in that, In step four, the parameters required for the interstellar fleet optimization algorithm include the total number of ships. M Maximum number of iterations tmax Number of iterations during the initial resource exploration phase n Periodic fleet reconfiguration strategy cycle m .

3. The novel intelligent fault diagnosis method for rotating machinery as described in claim 1, characterized in that, In step five, the CNN-LSTM neural network includes four convolutional layers, three pooling layers, one flattening layer, one LSTM layer, and one output layer. The shape of the CNN model is such that each of the first three convolutional layers is followed by a pooling layer, and the last convolutional layer is followed by a flattening layer. The data is then input into the LSTM model and then into the output layer. The activation function of the convolutional layers is the ReLU function, the stride is (2, 2), and zero padding is used. The pooling layer size is 2×2. The output layer outputs 4, the activation function is the softmax function, and the hyperparameters include the learning rate and L2 regularization parameter.

Images