CN115540935A - Novel method for diagnosing equipment fault of ship-protecting ballast water system - Google Patents
Novel method for diagnosing equipment fault of ship-protecting ballast water system Download PDFInfo
- Publication number
- CN115540935A CN115540935A CN202210552079.3A CN202210552079A CN115540935A CN 115540935 A CN115540935 A CN 115540935A CN 202210552079 A CN202210552079 A CN 202210552079A CN 115540935 A CN115540935 A CN 115540935A
- Authority
- CN
- China
- Prior art keywords
- neural network
- fault diagnosis
- data
- equipment
- ballast water
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01D—MEASURING NOT SPECIALLY ADAPTED FOR A SPECIFIC VARIABLE; ARRANGEMENTS FOR MEASURING TWO OR MORE VARIABLES NOT COVERED IN A SINGLE OTHER SUBCLASS; TARIFF METERING APPARATUS; MEASURING OR TESTING NOT OTHERWISE PROVIDED FOR
- G01D21/00—Measuring or testing not otherwise provided for
- G01D21/02—Measuring two or more variables by means not covered by a single other subclass
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Molecular Biology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Testing And Monitoring For Control Systems (AREA)
Abstract
The invention relates to a novel fault diagnosis method for a ship ballast water system device, which comprises the following steps: step 1: arranging a plurality of sensor monitoring points on the ship ballast water system equipment, and collecting characteristic parameters representing the running state of key equipment; and 2, step: constructing a fault diagnosis model consisting of a plurality of BP neural networks based on a cross entropy cost function, training and testing each BP neural network and acquiring a primary diagnosis result of equipment faults; and 3, step 3: and performing data fusion on the output result of each BP neural network by adopting a D-S evidence theory to obtain a final diagnosis result of the equipment fault. Compared with the prior art, the method has the advantages of reducing the time for delaying diagnosis, improving the precision of the fault diagnosis result, overcoming the defect that the fault cannot be accurately identified in real time by the conventional fault diagnosis method and the like.
Description
Technical Field
The invention relates to the technical field of fault diagnosis, in particular to a novel fault diagnosis method for a ship ballast water system device.
Background
The ballast water system consists of a plurality of devices, and plays a vital role in keeping the stable running of the naval vessel and ensuring the safety. It is known that the ballast water system equipment is prone to failure or damage in severe operating environments (such as extremely low or high temperature, corrosion, stormy weather, etc.) along with failures such as damage to sealing rings of the ballast water equipment, line failures of the ballast water equipment, blockage of pipelines of the ballast water system equipment, and the like during long-term navigation. Equipment failures tend to cause vessel maintenance, repair and navigation difficulties. In some cases, equipment failure can cause significant economic loss and even compromise the life safety of personnel on the vessel. Moreover, the arrangement of a plurality of devices on the ballast water system is relatively dispersed, which makes it difficult for workers to eliminate the hidden trouble. In addition, due to the complexity of the equipment structure, long-term ocean voyage of the naval vessel, extreme environment and the like, the possibility that multiple types of faults of different equipment occur simultaneously is increased, the signs of the similar faults may be different, and therefore the method for manually diagnosing the faults is invalid. Therefore, the intelligent fault diagnosis method for the ballast water system equipment is developed, the possible fault modes (the normal operation mode and the equipment fault mode) of the equipment can be diagnosed in real time, and the intelligent fault diagnosis method plays a vital role in reducing the risk of the failure of the naval vessel and ensuring the normal operation of the naval vessel.
The measurement data from the sensors may be inaccurate or incomplete due to factors such as sensor failure, bad weather, communication failure, and poor energy supply. It is thus known that the arrangement of a single sensor on a single device does not allow an accurate measurement of its operating state.
Disclosure of Invention
The invention aims to overcome the defects in the prior art and provide a novel fault diagnosis method for a ship ballast water system device.
The purpose of the invention can be realized by the following technical scheme:
a novel fault diagnosis method for a ship ballast water system device comprises the following steps:
step 1: arranging a plurality of sensor monitoring points on the ship ballast water system equipment, and collecting characteristic parameters representing the running state of key equipment;
and 2, step: constructing a fault diagnosis model consisting of a plurality of BP neural networks based on a cross entropy cost function, training and testing each BP neural network and acquiring a primary diagnosis result of equipment faults;
and step 3: and performing data fusion on the output result of each BP neural network by adopting a D-S evidence theory to obtain a final diagnosis result of the equipment fault.
In the step 1, the monitoring points of the sensor comprise a ballast water pump output power monitoring point, a plurality of flow monitoring points and a plurality of pressure monitoring points which are distributed on ballast water system equipment;
the characteristic parameters comprise pressure, flow and power;
key equipment of the ballast water system comprises a valve, a filter, a pump and a pipe section;
the operation states of key equipment of the ballast water system comprise pipe section leakage, valve blockage, ballast water pump shaft abrasion, seabed filter blockage and normal operation states.
In the step 1, the process of collecting the physical parameters of the fault characteristics representing the operation state of the key device specifically includes the following steps:
step 101: collecting fault test data through a plurality of sensor monitoring points distributed on the vessel ballast water system equipment;
step 102: uploading the fault measurement data to a computer, and performing feature extraction and feature selection operation on the fault measurement data by the computer to obtain feature data;
step 103: and carrying out normalization processing on the characteristic data, and dividing the characteristic data into training data for training the fault diagnosis model and test data for judging the diagnosis effect of the fault diagnosis model.
The fault diagnosis model comprises a first data preprocessing module, a second data preprocessing module and a decision fusion module:
the first data preprocessing module is used for preprocessing the characteristic data;
the second data preprocessing module comprises four BP neural networks with different network structures, extracted feature data are divided into four types based on the attributes of the feature data and the positions of sample points and are respectively used as the input of each BP neural network, the running state of ballast water system equipment is used as the output of each BP neural network, and the second data preprocessing module is used for carrying out normalization processing on the obtained fault diagnosis result;
and the decision fusion module is used for fusing the fault diagnosis results after the normalization processing to obtain a final fault diagnosis result.
Each BP neural network is a three-layer feedback network topological structure and comprises an input layer unit, a hidden layer unit and an output layer unit, all nodes of each layer unit are represented in a vector form, calculation is performed according to a vector operation method, and the BP neural networks adopt cross entropy cost functions to improve convergence speed and diagnosis precision and prevent phenomena of processing gradient disappearance and gradient explosion;
the input layer unit is provided with a plurality of input nodes, each input node corresponds to the characteristic data of the sensor monitoring point, and the input layer unit is activated by a sigmoid activation function;
the output layer unit is provided with a plurality of output nodes, and each output node corresponds to the preset running state of key equipment of the ballast water system;
and the hidden layer unit is activated by adopting a tanh activation function.
In the step 2, the process of training and testing each BP neural network and obtaining the initial diagnosis result of the equipment fault specifically comprises the following steps:
step 201: initializing and setting the characteristic parameters of each BP neural network, and classifying the training data and the test data after normalization processing according to the positions of the sample points and the attributes of the characteristic data;
step 202: inputting different types of training data into different BP neural networks for training;
step 203: and inputting test data to each trained BP neural network for testing to obtain a fault diagnosis result for representing the running state of the equipment, and analyzing the fault diagnosis result until the error meets the requirement.
In step 201, the characteristic parameters of each BP neural network include training times, learning rate, training error, input layer-to-intermediate layer weight, and intermediate layer-to-output layer weight.
In step 202, the process of inputting training data of different types into different BP neural networks for training specifically includes:
after selecting the characteristic parameters of each BP neural network, inputting the characteristic data into each BP neural network, adopting a cross entropy cost function as the cost function of each BP neural network, setting iteration times, obtaining various fault diagnosis results after the training of each BP neural network is completed, carrying out normalization processing on the fault diagnosis results of each BP neural network, and inputting the fault diagnosis results of each BP neural network after the normalization processing into each BP neural network for training until the iteration times are reached.
In step 203, the process of inputting the test data into each trained BP neural network for testing specifically includes:
the test data is input into each BP neural network for testing, most fault diagnosis results output by each BP neural network are processed and converted into 1-dimensional output results, namely the maximum probability value of each group of sample data is selected as an actual fault diagnosis result, and the expected result of the selected test data is compared with the actual fault diagnosis result output by each BP neural network to verify the equipment fault diagnosis performance of each BP neural network.
In the step 3, the process of performing data fusion on the output result of each BP neural network by using a D-S evidence theory specifically includes the following steps:
step 301: normalizing the fault diagnosis result of each BP neural network to obtain a reliability distribution result;
step 302: and fusing the confidence coefficient distribution results of each BP neural network by adopting a D-S evidence theory to obtain a final fault diagnosis result.
Compared with the prior art, the invention has the following beneficial effects:
1. according to the invention, a plurality of sensors are arranged on a single device to monitor the state of the device, so that the diagnosis precision of the neural network is improved;
2. according to the method, high-dimensional feature data are firstly divided into a plurality of low-dimensional feature data classes of different types according to the positions of sample points and the attributes of the feature data, then the feature data of different types are respectively used as the input of different BP neural networks according to classification results, and the feature data input by the different neural networks are classified based on the precision of the feature data, so that the problems of overlong diagnosis time, overfitting and inaccurate equipment fault classification results caused by the fact that a large amount of high-dimensional feature data are input into the same network are prevented;
3. the cross entropy cost function is used as the cost function of each BP neural network to avoid the problems of too slow training process and gradient reduction (such as sigmoids saturation and vanishing gradient) of each neural network, so that the fault diagnosis precision of a plurality of BP neural network devices is improved, and uncertainty and misdiagnosis rate are reduced;
4. in view of the advantages of the BP neural network based on the cross entropy cost function and the D-S evidence theory, the invention adopts a plurality of fault diagnosis algorithms combining the improved BP neural network based on the cross entropy cost function and the D-S evidence to improve the diagnosis precision of the plurality of BP neural networks.
Drawings
FIG. 1 is a flow chart of the method of the present invention.
FIG. 2 is a layout diagram of data monitoring points according to the present invention.
Fig. 3 is a schematic structural diagram of a fault diagnosis model of the present invention.
FIG. 4 is a schematic diagram of a single BP neural network structure according to the present invention.
Detailed Description
The invention is described in detail below with reference to the figures and specific embodiments. The present embodiment is implemented on the premise of the technical solution of the present invention, and a detailed implementation manner and a specific operation process are given, but the scope of the present invention is not limited to the following embodiments.
The invention provides a novel method for diagnosing equipment faults of a ship ballast water system, which is used for diagnosing the equipment faults of a ship ocean navigation process by combining a plurality of improved BP (back propagation) neural networks based on cross entropy cost functions and a D-S (Dempster-Shafer) evidence theory, firstly, arranging sensor monitoring points, collecting, extracting and selecting fault sample data, carrying out normalization processing on the fault sample data, taking 70% of fault sample data of a screen as training data to train a diagnostic model, taking the other 30% of the fault sample data as test data, and judging the diagnostic effect of the diagnostic model through the test data; secondly, initializing each BP neural network, classifying the normalized training data and the normalized test data according to the positions of the sample points and the attributes of the characteristic data, inputting the training data of different types into different neural networks for training, inputting the test data into each BP neural network after training for testing, and analyzing the output result; and finally, performing data fusion on the output result of each BP neural network according to a D-S evidence theory to obtain a final diagnosis result of the equipment fault.
Step 1: arranging monitoring points of a sensor of a ballast water system of an ocean naval vessel and acquiring data:
as shown in fig. 2, a plurality of sensors are arranged on each key device of the system to monitor the operating state of each device in real time, in this embodiment, 13 sensor monitoring points are arranged according to the distribution of a ballast water tank, an ultrasonic ultraviolet unit, a filter, a ballast water pump and an electric valve, including 1 ballast water pump output power sensor monitoring point Pm, 4 flow sensor monitoring points Q1, Q2, Q3 and Q4 and 8 pressure sensor monitoring points P1, P2, P3, P4, P5, P6, P7 and P8, and 5 device operating states are set as outputs, wherein the device operating states include pipe section leakage, valve blockage, ballast water pump shaft wear, seabed filter blockage and normal operating state;
the ship ballast water system is provided with a plurality of devices, a plurality of sensors are arranged on each device, each BP neural network represents the input of the failure characteristics of the device and comprises a plurality of different or same physical parameters, such as pressure, flow, power and the like, and the dimension of each physical parameter is different, for example: flow rate: 1212.5-1405.3m 3 H; power: 240.1-246.6kW; pressure: 0.01-1.66bar, the learning time of the BP neural network is increased by the feature data of different dimensions, and even the BP neural network cannot be converged, so that before the feature data is input into each BP neural network, normalization processing is required, the training time and the training error of each BP neural network are reduced, the convergence effect and the overall performance of each BP neural network are further improved, and the equipment fault diagnosis accuracy of each BP neural network is reduced.
And 2, step: establishing a fault diagnosis model consisting of a plurality of BP neural networks:
before designing each BP neural network, input and output of each BP neural network are determined, input of each BP neural network is a characteristic variable (characteristic data) of equipment faults, the input dimension of each BP neural network is consistent with the number of the characteristic variable of the equipment faults, the output dimension of each BP neural network is equal to the number of the equipment fault types, the acquired characteristic data are divided into two parts, namely training data and test data, output of the BP neural network comprises two parts, namely actual output and ideal output, the actual output is a probability value for representing the equipment faults output after each BP neural network is trained, the ideal output is a probability value for representing the equipment faults acquired actually, accuracy of equipment diagnosis is judged by comparing the actual output with the ideal output, output of each sample of each BP neural network is a probability value, the probability value range is a [0,1] range, and the probability value corresponding to a fault diagnosis result of jth sample data of an ith neural network is output as a maximum value of probabilities of all fault type probabilities in a jth group of characteristic data set of the ith neural network;
as shown in fig. 3, a fault diagnosis model based on a plurality of BP neural networks and a D-S evidence theory is established, the fault diagnosis model is composed of a plurality of BP neural networks, equipment fault symptom parameters, a D-S evidence theory fusion rule and a decision rule, the equipment fault symptom parameters are input into each BP neural network for primary diagnosis, the obtained fault diagnosis result is normalized, then the fault diagnosis result is classified and respectively input into BP neural networks of different network structures for training, and after the training is finished, test data is input into the fault diagnosis model for testing so as to verify the effectiveness of the fault diagnosis model.
Step 201: training and testing the fault diagnosis result:
initializing and setting parameters of each BP neural network:
as shown in FIG. 4, each BP neural network is composed of an input layer unit, a hidden layer unit and an output layer unit, where n is the number of neurons in the input layer, q is the number of neurons in the hidden layer, m is the number of neurons in the output layer, and x is the number of neurons in the output layer i (i =0,1, …, n-1) is the input variable for input layer neuron i (i =0,1, …, n-1), z j (j =0,1, …, q-1) is the output variable for hidden layer neuron j (j =0,1, …, q-1), y k (k =0,1, …, m-1) is the output variable for output layer neuron k (k =0,1, …, m-1), b j (j =0,1, …, q-1) is the bias variable for hidden layer neuron j, d k (k =0,1, …, m-1) is the bias variable for output layer neuron k (k =0,1, …, m-1), w ij (i =0,1, …, n-1, j =0,1, …, q-1) as an inputConnection weight variable, w-1, for layer neuron i (i =0,1, …, n-1) to hidden layer neuron j (j =0,1, …, q-1) jk (j =0,1, …, q-1, k =0,1, …, m-1) are the connection weight variables for hidden layer neuron j (j =0,1, …, q-1) and output layer neuron k (k =0,1, …, m-1).
Input layer unit design: as shown in fig. 3, in the present embodiment, 13 sensor monitoring points are arranged, and therefore, 13 network nodes are arranged on the input layer, which are respectively 1 ballast water pump output power sensor monitoring point Pm, 4 flow sensor monitoring points Q1, Q2, Q3 and Q4, and 8 pressure sensor monitoring points P1, P2, P3, P4, P5, P6, P7 and P8;
designing an output layer unit: as shown in fig. 3, the operation states of 5 kinds of equipment are defined in total, therefore, the output layer is provided with 5 network nodes, namely pipe section leakage, valve blockage, ballast water pump shaft abrasion, seabed filter blockage and normal operation state;
hidden layer unit design: the BP neural network has a theorem, that is, a single hidden layer BP neural network approximation can be adopted for any continuous function in a closed interval, based on which a 3-layer BP neural network can realize mapping of arbitrary n-dimensional data to an m-dimensional space, each BP neural network of the embodiment is designed with 1 hidden layer, generally, the number of hidden layer neurons is selected without a specific standard and needs to be designed through multiple experiments, and each BP neural network of the embodiment is designed with 7 hidden layer neurons.
And (3) designing an activation function: the activation functions of the common BP neural network comprise a Relu function, a tanh function and a sigmoid function, and from the requirement of problem solution, the activation functions used by an output layer and a hidden layer are determined according to the characteristics of the activation functions, wherein the output layer unit of the embodiment adopts the sigmoid activation function, and the hidden layer unit adopts the tanh activation function;
step 202: training each BP neural network:
in this embodiment, 13 sensor monitoring points are arranged in total, and 8 normal samples, plus 47 device failure samples, are selected from the collected measurement data, and 55 sample data are total, wherein 40 sample data are selected as training data of each BP neural network, and the remaining 15 sample data are selected as test data of each BP neural network;
before each BP neural network is trained, characteristic parameters (including training errors, training times, learning rate, input layer-hidden layer weight, hidden layer-output layer weight and the like) of each BP neural network are extracted:
the input layer-hidden layer weight and the hidden layer-output layer weight cannot be selected to be too large so as to prevent each BP neural network from entering saturation early and not achieving the preset effect, the selection range is usually between-1 and 1, the input layer-hidden layer weight of the embodiment is 0.7, and the hidden layer-output layer weight is 0.7;
the learning rate is too small, which affects the stability and convergence rate of each BP neural network, further affects the network error and increases the training time, while the learning rate is too large, which can reduce the time of sample training, but affects the overall stability of each neural network, generally, the value range of the learning rate is between 0.01 and 0.9, and the learning rate of the embodiment is 0.5;
after selecting the characteristic parameters of each BP neural network, inputting the characteristic data into each BP neural network, outputting the fault type of the equipment and the error of each BP neural network, if the accuracy of the output fault type of the equipment is low, fusing the output result of each BP neural network by adopting a D-S evidence theory, and in addition, if the iteration times are unreasonable, causing the network convergence speed to be slow, therefore, the embodiment takes the cross entropy cost function as the cost function of each BP neural network, sets the iteration times to 10000 times, writes a diagnosis program by adopting Python language, trains each BP neural network, obtains 5 fault diagnosis results after the training is finished, then normalizes the fault diagnosis result of each BP neural network, and finally inputs the fault diagnosis result of each BP neural network after the normalization processing into each BP neural network for training.
Step 203: testing each BP neural network:
in order to verify the equipment fault diagnosis performance of each trained BP neural network, each BP neural network needs to be tested, 15 pieces of test data are input into each BP neural network for testing, the fault diagnosis result output by each BP neural network is 5-dimensional, only one diagnosed equipment fault exists, on the basis, the fault diagnosis result needs to be processed, the 5-dimensional fault diagnosis result is converted into a 1-dimensional output result, namely, the maximum probability value of each group of sample data is selected as the actual fault diagnosis result, and finally, the expected result of the selected test data is compared with the actual fault diagnosis result output by each BP neural network so as to verify the equipment fault diagnosis performance of each BP neural network.
When the method adopts a plurality of BP neural networks to carry out fault diagnosis on equipment, each BP neural network selects the operation state of the ship ballast water system equipment corresponding to the maximum output probability as a diagnosis result.
The method adopts a plurality of filter inlet and outlet pressure signals, a plurality of pipe section pressure and flow signals, a pump inlet and outlet pressure signal and a pump shaft power signal to carry out characteristic preprocessing on the operation state of the ship ballast water system equipment of the ship, combines a plurality of improved BP neural networks based on cross entropy cost functions and a D-S evidence theory, and realizes accurate diagnosis of different fault modes.
The foregoing detailed description of the preferred embodiments of the invention has been presented. It should be understood that numerous modifications and variations could be devised by those skilled in the art in light of the present teachings without departing from the inventive concepts. Therefore, the technical solutions available to those skilled in the art through logic analysis, reasoning and limited experiments based on the prior art according to the concept of the present invention should be within the scope of protection defined by the claims.
Claims (10)
1. A novel fault diagnosis method for ship ballast water system equipment is characterized by comprising the following steps:
step 1: arranging a plurality of sensor monitoring points on the ship ballast water system equipment, and collecting characteristic parameters representing the running state of key equipment;
and 2, step: constructing a fault diagnosis model consisting of a plurality of BP neural networks based on a cross entropy cost function, training and testing each BP neural network and acquiring a primary diagnosis result of equipment faults;
and step 3: and performing data fusion on the output result of each BP neural network by adopting a D-S evidence theory to obtain a final diagnosis result of the equipment fault.
2. The fault diagnosis method for the novel ship ballast water system equipment in the step 1 is characterized in that the sensor monitoring points comprise ballast water pump output power monitoring points, a plurality of flow monitoring points and a plurality of pressure monitoring points which are distributed on the ballast water system equipment;
the characteristic parameters comprise pressure, flow and power;
the ballast water system key equipment comprises a valve, a filter, a pump and a pipe section;
the operation states of key equipment of the ballast water system comprise pipe section leakage, valve blockage, ballast water pump shaft abrasion, seabed filter blockage and normal operation states.
3. The novel method for diagnosing the equipment fault of the ship ballast water system according to claim 1, wherein the step 1 of collecting the physical parameters representing the fault characteristics of the running state of the key equipment specifically comprises the following steps:
step 101: collecting fault test data through a plurality of sensor monitoring points distributed on the vessel ballast water system equipment;
step 102: uploading the fault measurement data to a computer, and performing feature extraction and feature selection operation on the fault measurement data by the computer to obtain feature data;
step 103: and carrying out normalization processing on the characteristic data, and dividing the characteristic data into training data for training the fault diagnosis model and test data for judging the diagnosis effect of the fault diagnosis model.
4. The novel fault diagnosis method for the ship ballast water system equipment of the ship as claimed in claim 3, wherein the fault diagnosis model comprises a first data preprocessing module, a second data preprocessing module and a decision fusion module:
the first data preprocessing module is used for preprocessing the characteristic data;
the second data preprocessing module comprises four BP neural networks with different network structures, extracted feature data are divided into four types based on the attributes of the feature data and the positions of sample points and are respectively used as the input of each BP neural network, the running state of ballast water system equipment is used as the output of each BP neural network, and the second data preprocessing module is used for carrying out normalization processing on the obtained fault diagnosis result;
and the decision fusion module is used for fusing the fault diagnosis results after the normalization processing to obtain a final fault diagnosis result.
5. The method for diagnosing the equipment faults of the novel ship ballast water system according to claim 4, wherein each BP neural network is of a three-layer feedback network topology structure which comprises an input layer unit, an implicit layer unit and an output layer unit, all nodes of each layer unit are represented in a vector form and are calculated according to a vector operation method, and the BP neural network adopts a cross entropy cost function to improve convergence speed and diagnosis accuracy and prevent the phenomena of processing gradient disappearance and gradient explosion;
the input layer unit is provided with a plurality of input nodes, each input node corresponds to the characteristic data of the sensor monitoring point, and the input layer unit is activated by a sigmoid activation function;
the output layer unit is provided with a plurality of output nodes, and each output node corresponds to the preset running state of key equipment of the ballast water system;
the hidden layer unit is activated by adopting a tanh activation function.
6. The method for diagnosing equipment faults of the novel ship ballast water system according to claim 1, wherein in the step 2, the process of training and testing each BP neural network and obtaining the initial diagnosis result of the equipment faults specifically comprises the following steps:
step 201: initializing and setting the characteristic parameters of each BP neural network, and classifying the training data and the test data after normalization processing according to the positions of the sample points and the attributes of the characteristic data;
step 202: inputting different types of training data into different BP neural networks for training;
step 203: and inputting the test data into each trained BP neural network for testing to obtain a fault diagnosis result for representing the running state of the equipment, and analyzing the fault diagnosis result until the error meets the requirement.
7. The method as claimed in claim 6, wherein in step 201, the characteristic parameters of each BP neural network include training times, learning rate, training error, weight from input layer to intermediate layer, and weight from intermediate layer to output layer.
8. The method as claimed in claim 6, wherein in step 202, the process of inputting different types of training data into different BP neural networks for training specifically comprises:
after selecting the characteristic parameters of each BP neural network, inputting the characteristic data into each BP neural network, adopting a cross entropy cost function as the cost function of each BP neural network, setting iteration times, obtaining various fault diagnosis results after each BP neural network is trained, normalizing the fault diagnosis results of each BP neural network, and inputting the fault diagnosis results of each BP neural network after the normalization processing into each BP neural network for training until the iteration times are reached.
9. The method as claimed in claim 6, wherein the step 203 of inputting the test data into each trained BP neural network for testing specifically comprises:
the test data is input into each BP neural network for testing, most fault diagnosis results output by each BP neural network are processed and converted into 1-dimensional output results, namely the maximum probability value of each group of sample data is selected as an actual fault diagnosis result, and the expected result of the selected test data is compared with the actual fault diagnosis result output by each BP neural network to verify the equipment fault diagnosis performance of each BP neural network.
10. The novel fault diagnosis method for the ship ballast water system equipment in the step 3 is characterized in that the process of performing data fusion on the output result of each BP neural network by adopting the D-S evidence theory specifically comprises the following steps:
step 301: normalizing the fault diagnosis result of each BP neural network to obtain a reliability distribution result;
step 302: and fusing the confidence coefficient distribution results of each BP neural network by adopting a D-S evidence theory to obtain a final fault diagnosis result.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210552079.3A CN115540935A (en) | 2022-05-18 | 2022-05-18 | Novel method for diagnosing equipment fault of ship-protecting ballast water system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210552079.3A CN115540935A (en) | 2022-05-18 | 2022-05-18 | Novel method for diagnosing equipment fault of ship-protecting ballast water system |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115540935A true CN115540935A (en) | 2022-12-30 |
Family
ID=84723546
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210552079.3A Pending CN115540935A (en) | 2022-05-18 | 2022-05-18 | Novel method for diagnosing equipment fault of ship-protecting ballast water system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115540935A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117176550A (en) * | 2023-09-25 | 2023-12-05 | 云念软件(广东)有限公司 | Integrated operation maintenance method and system based on fault identification |
-
2022
- 2022-05-18 CN CN202210552079.3A patent/CN115540935A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117176550A (en) * | 2023-09-25 | 2023-12-05 | 云念软件(广东)有限公司 | Integrated operation maintenance method and system based on fault identification |
| CN117176550B (en) * | 2023-09-25 | 2024-03-19 | 云念软件(广东)有限公司 | Integrated operation maintenance method and system based on fault identification |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109141847B (en) | Aircraft system fault diagnosis method based on MSCNN deep learning | |
| CN110288046B (en) | Fault prediction method based on wavelet neural network and hidden Markov model | |
| CN108256173B (en) | Gas circuit fault diagnosis method and system for dynamic process of aircraft engine | |
| CN112016251B (en) | Nuclear power device fault diagnosis method and system | |
| CN111680875B (en) | Unmanned aerial vehicle state risk fuzzy comprehensive evaluation method based on probability baseline model | |
| CN112199890B (en) | System-level fault diagnosis method for integrated nuclear power device | |
| CN109298633A (en) | Chemical production process fault monitoring method based on adaptive piecemeal Non-negative Matrix Factorization | |
| WO2021114320A1 (en) | Wastewater treatment process fault monitoring method using oica-rnn fusion model | |
| WO2019178930A1 (en) | Fault diagnosis method for mechanical device | |
| CN116384224A (en) | Aero-engine life prediction method based on conditional parameter dynamic convolutional neural network | |
| CN115540935A (en) | Novel method for diagnosing equipment fault of ship-protecting ballast water system | |
| CN116595319A (en) | Prediction method and system applied to rail transit motor health state evaluation | |
| CN114548154A (en) | Intelligent diagnosis method and device for important service water pump | |
| CN115791174B (en) | Rolling bearing abnormality diagnosis method, system, electronic equipment and storage medium | |
| CN105279553B (en) | A kind of height adds to water system fault degree recognition methods | |
| CN116842379A (en) | Mechanical bearing residual service life prediction method based on DRSN-CS and BiGRU+MLP models | |
| CN111319731A (en) | Fault diagnosis method for ballast water equipment | |
| CN116578922A (en) | Valve cooling system fault diagnosis method and device based on multichannel convolutional neural network | |
| CN115204272A (en) | Industrial system fault diagnosis method and equipment based on multi-sampling rate data | |
| CN111061191B (en) | Distributed oil-gas storage tank remote operation and maintenance method | |
| Vladov et al. | Modified Neural Network Method for Diagnostics the Helicopters Turboshaft Engines Operational Status at Flight Modes | |
| CN112560252A (en) | Prediction method for residual life of aircraft engine | |
| CN113158537A (en) | Aeroengine gas circuit fault diagnosis method based on LSTM combined attention mechanism | |
| Wang et al. | Complex equipment diagnostic reasoning based on neural network algorithm | |
| Vychuzhanin et al. | Intellectualization Method and Model of Complex Technical System's Failures Risk Estimation and Prediction. |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |