Marine steam turbine fault diagnosis method

Abstract

A ship steam turbine fault diagnosis method relates to the field of ship steam turbine fault diagnosis. The invention aims to solve the problem that smoothing factors are difficult to determine when ship steam turbine fault diagnosis is carried out by utilizing a probabilistic neural network. According to the invention, the fault feature data of the steam turbine is collected, and the fault feature data is input into the probabilistic neural network fault diagnosis model to obtain a fault identification result. The steam turbine data is processed by adopting a Min-Max standardization method, so that the distribution and relative relation of the original data can be reserved, the scale consistency of different characteristics can be ensured, and the reliability of diagnostic data can be effectively ensured. The probabilistic neural network is optimized through a salat swarm intelligent optimization algorithm, an optimal smoothing factor can be obtained, and the problem of smoothing factor selection is solved. The method has the advantages of being easy to train, rapid in convergence, high in accuracy and the like, and faults can be rapidly and effectively recognized.

Description

Technical Field

[0001] The present invention belongs to the field of fault diagnosis of marine steam turbines. Background Art

[0002] As the main equipment for the power of steam ships, once a steam turbine fails and the failure cannot be diagnosed and eliminated in time, it will cause significant economic and property losses. In serious cases, safety accidents will occur, and the consequences will be unimaginable. Carrying out fault diagnosis of marine steam turbines and eliminating faults in time is of great significance to the safe navigation of ships. Due to the complex operation mechanism, high precision, and many uncertain factors in the on-site environment of marine steam turbines, and various factors affecting the occurrence of faults, it is very difficult to diagnose the faults of steam turbines in a timely and accurate manner according to the fault characteristics of the steam turbines. Therefore, quickly and accurately diagnosing the faults of marine steam turbines has become a popular problem that needs to be solved urgently.

[0003] At present, the traditional fault diagnosis methods for marine steam turbines mainly rely on experience, that is, judging faults by analyzing the operation mechanism of marine steam turbines. This method is often restricted by engineering practices such as insufficient experience and difficulties in disassembling marine steam turbines due to limited space on ships.

[0004] With the development of neural network technology, especially the probabilistic neural network (PNN) as a kind of neural network model, the concept of probability theory is introduced in the model design and training process, making the model more flexible and able to adapt to different data distributions and application scenarios, thus being applicable to various complex problems such as fault diagnosis of marine steam turbines. However, the probabilistic neural network has the problem that it is difficult to determine the smoothing factor. If the value of the smoothing factor is larger, the approximation error will increase accordingly; on the contrary, if the value of the smoothing factor is smaller, it is easy to cause overfitting. Therefore, it is more important to reasonably select the smoothing factor for the probabilistic neural network. Summary of the Invention

[0005] The present invention is to solve the problem that it is difficult to determine the smoothing factor when using the probabilistic neural network for fault diagnosis of marine steam turbines. Now, a fault diagnosis method for marine steam turbines that optimizes the probabilistic neural network model with the sand cat swarm algorithm (SCSO) is provided. The sand cat swarm algorithm optimizes the probabilistic neural network model, which can improve the performance of the probabilistic neural network and extract the most suitable smoothing factor, making it more applicable to the fault diagnosis of marine steam turbines, and then effectively overcoming the limitations of the probabilistic neural network, providing technical support and decision-making support for the fault diagnosis of marine steam turbines.

[0006] A fault diagnosis method for marine steam turbines includes:

[0007] Collecting the fault feature data of the steam turbine and inputting the fault feature data into the probabilistic neural network fault diagnosis model to obtain a fault recognition result.

[0008] Further, the above probability neural network fault diagnosis model includes: an input layer, a pattern layer, a summation layer, and an output layer;

[0009] The input layer is used to receive the fault feature data and transfer it to the pattern layer;

[0010] The pattern layer is used to perform Gaussian function calculation on the fault feature data to obtain the neuron category to which the fault feature data belongs;

[0011] The summation layer is used to take the average of the outputs of neurons of the same category to obtain the estimated probability of each category of neurons;

[0012] The output layer is used to output the fault category with the highest estimated probability as the fault recognition result.

[0013] Further, the above Gaussian function expression is:

[0014]

[0015] where σ is the smoothing factor, p is the dimension of the fault feature data, x ij is the i-th neuron in the pattern layer of the j-th class, and Φ ij (·) represents the Gaussian function.

[0016] Further, the training process of the above probability neural network fault diagnosis model includes:

[0017] Collect the fault feature data and fault categories of the steam turbine;

[0018] Optimize the smoothing factor of the probability neural network fault diagnosis model based on the sand cat swarm optimization algorithm to obtain a trained probability neural network fault diagnosis model.

[0019] Further, the optimization of the smoothing factor of the probability neural network fault diagnosis model based on the sand cat swarm optimization algorithm includes:

[0020] Traverse each sand cat individual and calculate the optimal candidate position of each sand cat individual;

[0021] In the sand cat swarm, take the optimal candidate position with the smallest fitness value as the optimal smoothing factor of the probability neural network fault diagnosis model, so as to obtain the probability neural network fault diagnosis model optimized by the sand cat swarm optimization algorithm.

[0022] Further, the calculation of the optimal candidate position of each sand cat individual includes:

[0023] Initialize the initial position of the sand cat swarm;

[0024] S1: Update the position of the sand cat swarm individuals according to the following formula:

[0025]

[0026] where \(t = 0, 1, 2, \cdots, T\), \(T\) is the maximum number of iterations, \(n = 1, 2, \cdots, N\), \(d = 1, 2, \cdots, D\), \(N\) is the population size, \(D\) is the dimension of variables, \(P\) n,d (t + 1) is the position of the \(d\)-th dimension fennec fox individual in the \(n\)-th population at the \((t + 1)\)-th iteration, \(P\) n,d (t) is the position of the \(d\)-th dimension fennec fox individual in the \(n\)-th population at the \(t\)-th iteration, \(P\) b is the optimal candidate position, \(P\) r is the random position, \(\theta\) is the random angle, \(r\) represents the sensitivity range of the fennec fox, \(rand\) is a random number between \(0\) and \(1\), \(R\) is a parameter for choosing to use the search behavior or the attack behavior, and \(R = 2r\) G rand - r G , S is a constant;

[0027] S2: Take \(P\) n,d (t + 1) as the smoothing factor of the probabilistic neural network fault diagnosis model, input the collected fault feature data into the probabilistic neural network fault diagnosis model to obtain the fault type recognition result, and calculate the model fitness value corresponding to \(P\) n,d (t + 1);

[0028] S3: Judge whether the model fitness value is less than the minimum fitness value. If so, take the model fitness value as the minimum fitness value and take \(P\) n,d (t + 1) as the optimal candidate position \(P\) b , otherwise do not update \(F\) min and \(P\) b ;

[0029] S4: Judge whether \(t + 2>T\) holds. If so, take the optimal candidate position \(P\) obtained in S3 b as the optimal candidate position of the \(d\)-th dimension fennec fox individual in the \(n\)-th population, otherwise set \(t=t + 1\) and return to S1.

[0030] Furthermore, the expression of the above model fitness value is:

[0031]

[0032] where \(F\) is the model fitness value, \(M\) is the number of fault feature data, and \(Z\) is the number of accurately recognized ones.

[0033] Furthermore, the expression of the initial position \(P(0)\) of the above fennec fox swarm is:

[0034]

[0035] P N,D (0) represents the initial position of the D-th dimensional sand cat individual in the N-th population.

[0036] Furthermore, after collecting the fault feature data and fault categories of the steam turbine, the fault feature data is standardized and then the smoothing factor is optimized.

[0037] Furthermore, the standardization processing of the fault feature data includes:

[0038] Use the Min-Max standardization method for standardization processing:

[0039]

[0040] where x s is the standardized fault feature data, x is the fault feature data before standardization, x min is the minimum value of the fault feature data, x max is the maximum value of the fault feature data.

[0041] The advantages of the present invention are as follows:

[0042] 1. The present invention uses the Min-Max standardization method to process the steam turbine data, which can not only retain the distribution and relative relationship of the original data, but also ensure the scale consistency of different features, and can effectively ensure the reliability of the diagnostic data.

[0043] 2. The present invention optimizes the probabilistic neural network through the sand cat swarm intelligent optimization algorithm, and can obtain the optimal smoothing factor, which can effectively solve the problem of selecting the smoothing factor.

[0044] 3. The present invention has the advantages of easy training, fast convergence, high accuracy, etc., and can quickly and effectively identify faults. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 is the flow chart of the present invention;

[0046] Figure 2 is the structure diagram of the probabilistic neural network. DETAILED DESCRIPTION OF THE INVENTION

[0047] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts belong to the scope of protection of the present invention. It should be noted that, without conflict, the embodiments in the present invention and the features in the embodiments may be combined with each other.

[0048] Refer to Figure 1 and Figure 2 Specifically describe this embodiment. A method for diagnosing faults in a marine steam turbine described in this embodiment includes:

[0049] Step 1, collect steam turbine data to obtain fault feature data and fault categories.

[0050] The fault feature data includes: fault feature data 1, fault feature data 2,..., fault feature data m;

[0051] The fault categories include: fault category 1, fault category 2,..., fault category n.

[0052] Step 2, process the fault feature data using a standardization method to obtain the fault feature data.

[0053] The standardization method is the Min - Max standardization method, and the formula is as follows:

[0054]

[0055] where, x s is the standardized fault feature data, x is the fault feature data before standardization, x min is the minimum value of the fault feature data, x max is the maximum value of the fault feature data.

[0056] Step 3, divide the fault feature data into training data and test data.

[0057] Divide X% of the fault feature data into training data, and divide Y% of the fault feature data into test data, where X + Y = 100.

[0058] Step 4, construct a probabilistic neural network fault diagnosis model, use the fault feature data as the input of the probabilistic neural network fault diagnosis model, and use the fault category as the output of the probabilistic neural network fault diagnosis model.

[0059] The probabilistic neural network fault diagnosis model includes: an input layer, a pattern layer, a summation layer, and an output layer.

[0060] The input layer is used to receive fault feature data and transfer it to the pattern layer.

[0061] The pattern layer is used to perform Gaussian function calculation on the fault feature data of the input layer. The formula is as follows:

[0062]

[0063] Among them, σ is the smoothing factor, p is the dimension of the fault feature data, and x ij is the i-th pattern layer neuron of the j-th class, and Φ ij (·) represents the Gaussian function.

[0064] The summation layer is used to take the average of the outputs of neurons of the same category to obtain the estimated probabilities of neurons of each category.

[0065] The output layer outputs the fault category with the highest estimated probability as the fault recognition result.

[0066] Step 5: Optimize the probability neural network fault diagnosis model with the sand cat swarm optimization algorithm, and use the training data to train to obtain the optimal smoothing factor. The specific steps are as follows:

[0067] Initialize the sand cat swarm: The initial position P(0) of the sand cat swarm is expressed as:

[0068]

[0069] Among them, N is the population size, D is the variable dimension, and P N,D (0) represents the initial position of the d-th dimensional sand cat individual in the N-th population.

[0070] Traverse each sand cat individual and calculate the optimal candidate position of each sand cat individual. In the sand cat swarm, the optimal candidate position with the smallest fitness value is used as the optimal position, and this optimal position is the optimal smoothing factor of the probability neural network fault diagnosis model, so as to obtain the probability neural network fault diagnosis model optimized by the sand cat swarm optimization algorithm.

[0071] The steps of calculating the optimal candidate position corresponding to each sand cat individual include:

[0072] S1: Update the position of the sand cat swarm individuals according to the following formula:

[0073]

[0074] Among them, t = 0, 1, 2,..., T, and T is the maximum number of iterations;

[0075] P n,d (t + 1) is the position of the d-th dimensional sand cat individual in the n-th population at the (t + 1)-th iteration, θ ∈ [0, 360] is a random angle, and P bis the candidate position, P r is the random position, P n,d (t) is the position of the d-th dimensional sand cat individual in the n-th population at the t-th iteration;

[0076] r represents the sensitivity range of each sand cat and r = r G rand, r G is a parameter and linearly decreases from 2 to 0 as the number of iterations increases, and S is a constant with a value of 2, rand is a random number between 0 and 1;

[0077] R is a parameter for choosing to use the search behavior or the attack behavior, and R = 2r G rand - r G , if |R| > 1, then update the position of the next generation according to the search state, if |R| ≤ 1, then update the position of the next generation according to the attack state.

[0078] S2: Take P n,d (t + 1) as the smoothing factor of the PNN model, input the fault feature data into the PNN model, obtain the fault type recognition result, and then calculate P n,d (t + 1) corresponding fitness value F:

[0079]

[0080] where, M is the number of fault feature data, and Z is the number of accurately recognized ones.

[0081] S3: Judge whether the fitness value F is less than the minimum fitness value F min , if so, take the fitness value F as the minimum fitness value F min , and take P n,d (t + 1) as the optimal candidate position P b , otherwise do not update F min and P b .

[0082] S4: Judge whether t + 2 > T holds, if so, take the optimal candidate position P obtained in S3 b as the optimal candidate position of the d-th dimensional sand cat individual in the n-th population, otherwise make t = t + 1 and return to S1.

[0083] Step 6, input the test data into the probability neural network fault diagnosis model optimized by the sand cat swarm optimization algorithm, and output the fault diagnosis result. Specific embodiments

[0085] Fault diagnosis is now carried out on a marine steam turbine. The steam turbine data includes the steam turbine speed, steam turbine power, steam turbine torque, steam turbine bearing temperature, and steam turbine vibration frequency. The fault categories include bearing damage, blade corrosion, self-excited vibration, seal looseness, and misalignment.

[0086] 70% of the fault feature data is divided into training data, and 30% of the fault feature data is divided into test data.

[0087] Initialize the sand cat swarm: the population size N = 20, the variable dimension D = 1, and the maximum number of iterations T = 50.

[0088] The initial position of the sand cat swarm is represented as:

[0089]

[0090] The method described in the specific implementation manner is tested with the steam turbine data, and the fault diagnosis accuracy rate is as high as 96.33%.

[0091] In summary, the present invention discloses a fault diagnosis method for a marine steam turbine aiming at the problem of fault diagnosis of a marine steam turbine. The main features of the present invention are: collecting steam turbine data to obtain fault feature data and fault categories; processing the fault feature data with a standardization method to obtain fault feature data; dividing the fault feature data into training data and test data; constructing a probabilistic neural network fault diagnosis model, using the fault feature data as the input of the probabilistic neural network fault diagnosis model, and using the fault categories as the output of the probabilistic neural network fault diagnosis model; optimizing the probabilistic neural network fault diagnosis model with a sand cat swarm optimization algorithm, and training with the training data to obtain an optimal smoothing factor; inputting the test data into the probabilistic neural network fault diagnosis model optimized by the sand cat swarm optimization algorithm, and outputting the fault diagnosis result. The present invention has the advantages of easy training, fast convergence, high accuracy, etc., and can quickly and effectively identify faults.

[0092] Although the present invention has been described herein with reference to specific embodiments, it should be understood that these embodiments are merely examples of the principles and applications of the present invention. Therefore, it should be understood that many modifications can be made to the exemplary embodiments, and other arrangements can be designed, as long as they do not deviate from the spirit and scope of the present invention as defined by the appended claims. It should be understood that different dependent claims and the features described herein can be combined in a manner different from that described in the original claims. It should also be understood that the features described in connection with a single embodiment can be used in other described embodiments.

Claims

1. A method for diagnosing faults of a ship steam turbine, characterized in that: include: The fault characteristic data of the steam turbine is collected and input into the probabilistic neural network fault diagnosis model to obtain the fault identification result.

2. A method for diagnosing faults of a ship steam turbine according to claim 1, characterized in that: The probabilistic neural network fault diagnosis model includes: an input layer, a pattern layer, a summation layer and an output layer; The input layer is used to receive the fault characteristic data and pass it to the mode layer; The pattern layer is used to perform Gaussian function calculation on the fault feature data to obtain the neuron category to which the fault feature data belongs; The summation layer is used to average the outputs of neurons of the same category to obtain the estimated probability of neurons of each category; The output layer is used to output the fault category with the highest estimated probability as the fault identification result.

3. A method for diagnosing faults of a ship steam turbine according to claim 2, characterized in that: The Gaussian function expression is: Among them, σ is the smoothing factor, p is the dimension of fault feature data, x ij is the i-th pattern layer neuron of the j-th class, Φ ij (·) represents a Gaussian function.

4. A method for diagnosing faults of a ship steam turbine according to claim 1, characterized in that: The training process of the probabilistic neural network fault diagnosis model includes: Collect fault characteristic data and fault categories of steam turbines; The smoothing factor of the probabilistic neural network fault diagnosis model is optimized based on the sand cat swarm optimization algorithm to obtain a trained probabilistic neural network fault diagnosis model.

5. A method for diagnosing faults of a ship steam turbine according to claim 4, characterized in that: The method of optimizing the smoothing factor of the probabilistic neural network fault diagnosis model based on the sand cat swarm optimization algorithm includes: Traverse each sand cat individual and calculate the optimal candidate position of each sand cat individual; In the sand cat swarm, the optimal candidate position with the smallest fitness value is taken as the optimal smoothing factor of the probabilistic neural network fault diagnosis model, so as to obtain the probabilistic neural network fault diagnosis model optimized by the sand cat swarm optimization algorithm.

6. A method for diagnosing faults of a ship steam turbine according to claim 5, characterized in that: The step of calculating the optimal candidate position of each sand cat individual comprises: Initialize the initial position of the sand cat group; S1: Update the individual positions of the sand cat group according to the following formula: Where t=0,1,2,...,T, T is the maximum number of iterations, n=1,2,...,N, d=1,2,...,D, N is the population size, D is the variable dimension, P n,d (t+1) is the position of the dth individual of the Visha cat in the nth population at the t+1th iteration, P n,d (t) is the position of the d-th Visha cat individual in the n-th population at the t-th iteration, P b is the optimal candidate position, P r is a random position, θ is a random angle, r represents the sensitivity range of the sand cat, rand is a random number between 0 and 1, R is a parameter for selecting the search behavior or the attack behavior, and R = 2r G rand-r G , S is a constant; S2: P n,d (t+1) is used as the smoothing factor of the probabilistic neural network fault diagnosis model, and the collected fault feature data is input into the probabilistic neural network fault diagnosis model to obtain the fault type recognition result, and the fault type recognition result is used to calculate P n,d (t+1) corresponds to the model fitness value; S3: Determine whether the model fitness value is less than the minimum fitness value. If so, take the model fitness value as the minimum fitness value and set P n,d (t+1) is the optimal candidate position P b Otherwise, F min and P b Make updates; S4: Determine whether t+2>T is true, if so, the optimal candidate position P obtained in S3 is b As the optimal candidate position of the d-th Visha cat individual in the n-th population, otherwise set t=t+1 and return to S1.

7. A method for diagnosing faults of a marine steam turbine according to claim 6, characterized in that: The expression of the model fitness value is: Among them, F is the model fitness value, M is the number of fault feature data, and Z is the number of accurate identifications.

8. A method for diagnosing faults of a marine steam turbine according to claim 6, characterized in that: The initial position P(0) of the sand cat group is expressed as: P N,D (0) represents the initial position of the Dth Visha cat individual in the Nth population.

9. A method for diagnosing faults of a marine steam turbine according to claim 4, characterized in that: After the fault characteristic data and fault categories of the steam turbine are collected, the fault characteristic data are standardized and then the smoothing factor is optimized.

10. A method for diagnosing faults of a ship steam turbine according to claim 9, characterized in that: The standardization process of the fault characteristic data includes: The Min-Max standardization method is used for standardization: Among them, x s is the fault characteristic data after standardization, x is the fault characteristic data before standardization, and x min is the minimum value of the fault characteristic data, x max is the maximum value of the fault characteristic data.

Images