CN116632295A - Multi-fault diagnosis method based on SOFC system - Google Patents
Multi-fault diagnosis method based on SOFC system Download PDFInfo
- Publication number
- CN116632295A CN116632295A CN202310732313.5A CN202310732313A CN116632295A CN 116632295 A CN116632295 A CN 116632295A CN 202310732313 A CN202310732313 A CN 202310732313A CN 116632295 A CN116632295 A CN 116632295A
- Authority
- CN
- China
- Prior art keywords
- fault
- state
- representing
- sofc
- capsule
- 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
- 238000003745 diagnosis Methods 0.000 title claims abstract description 56
- 238000000034 method Methods 0.000 title claims abstract description 31
- 239000002775 capsule Substances 0.000 claims abstract description 56
- 239000002131 composite material Substances 0.000 claims abstract description 20
- 239000000446 fuel Substances 0.000 claims abstract description 17
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 6
- 238000010206 sensitivity analysis Methods 0.000 claims abstract description 5
- 230000035945 sensitivity Effects 0.000 claims abstract description 4
- 238000000605 extraction Methods 0.000 claims description 26
- 230000006870 function Effects 0.000 claims description 21
- 239000011159 matrix material Substances 0.000 claims description 17
- 238000011176 pooling Methods 0.000 claims description 15
- 238000012549 training Methods 0.000 claims description 14
- 230000008569 process Effects 0.000 claims description 9
- 238000012937 correction Methods 0.000 claims description 7
- 230000014509 gene expression Effects 0.000 claims description 7
- 238000007781 pre-processing Methods 0.000 claims description 7
- 238000013135 deep learning Methods 0.000 claims description 6
- PXFBZOLANLWPMH-UHFFFAOYSA-N 16-Epiaffinine Natural products C1C(C2=CC=CC=C2N2)=C2C(=O)CC2C(=CC)CN(C)C1C2CO PXFBZOLANLWPMH-UHFFFAOYSA-N 0.000 claims description 5
- 230000002093 peripheral effect Effects 0.000 claims description 5
- 239000007787 solid Substances 0.000 claims description 5
- 230000009466 transformation Effects 0.000 claims description 5
- 238000013527 convolutional neural network Methods 0.000 claims description 4
- 238000002347 injection Methods 0.000 claims description 3
- 239000007924 injection Substances 0.000 claims description 3
- 230000010354 integration Effects 0.000 claims description 3
- 230000010287 polarization Effects 0.000 claims description 3
- 238000013461 design Methods 0.000 claims description 2
- 238000002405 diagnostic procedure Methods 0.000 claims description 2
- 230000000694 effects Effects 0.000 abstract description 2
- 239000007789 gas Substances 0.000 description 24
- 239000001257 hydrogen Substances 0.000 description 17
- 229910052739 hydrogen Inorganic materials 0.000 description 17
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 12
- 229910001868 water Inorganic materials 0.000 description 12
- QVGXLLKOCUKJST-UHFFFAOYSA-N atomic oxygen Chemical compound [O] QVGXLLKOCUKJST-UHFFFAOYSA-N 0.000 description 11
- 239000001301 oxygen Substances 0.000 description 11
- 229910052760 oxygen Inorganic materials 0.000 description 11
- UFHFLCQGNIYNRP-UHFFFAOYSA-N Hydrogen Chemical compound [H][H] UFHFLCQGNIYNRP-UHFFFAOYSA-N 0.000 description 9
- 230000004913 activation Effects 0.000 description 9
- 150000002431 hydrogen Chemical class 0.000 description 8
- 239000013598 vector Substances 0.000 description 7
- 238000006243 chemical reaction Methods 0.000 description 6
- 238000012360 testing method Methods 0.000 description 6
- 238000004364 calculation method Methods 0.000 description 5
- 238000010586 diagram Methods 0.000 description 5
- 238000013507 mapping Methods 0.000 description 5
- 238000010606 normalization Methods 0.000 description 4
- 239000000243 solution Substances 0.000 description 4
- 210000004027 cell Anatomy 0.000 description 3
- 230000008859 change Effects 0.000 description 3
- 238000003487 electrochemical reaction Methods 0.000 description 3
- 230000004927 fusion Effects 0.000 description 3
- 238000013178 mathematical model Methods 0.000 description 3
- 238000011160 research Methods 0.000 description 3
- 238000012795 verification Methods 0.000 description 3
- IJGRMHOSHXDMSA-UHFFFAOYSA-N Atomic nitrogen Chemical compound N#N IJGRMHOSHXDMSA-UHFFFAOYSA-N 0.000 description 2
- 206010044565 Tremor Diseases 0.000 description 2
- 230000006399 behavior Effects 0.000 description 2
- 230000008878 coupling Effects 0.000 description 2
- 238000010168 coupling process Methods 0.000 description 2
- 238000005859 coupling reaction Methods 0.000 description 2
- 230000001419 dependent effect Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 210000002569 neuron Anatomy 0.000 description 2
- 238000005457 optimization Methods 0.000 description 2
- 238000010248 power generation Methods 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 230000004044 response Effects 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 230000000295 complement effect Effects 0.000 description 1
- 150000001875 compounds Chemical class 0.000 description 1
- 230000006835 compression Effects 0.000 description 1
- 238000007906 compression Methods 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 230000002950 deficient Effects 0.000 description 1
- 238000009792 diffusion process Methods 0.000 description 1
- 239000003792 electrolyte Substances 0.000 description 1
- 238000003912 environmental pollution Methods 0.000 description 1
- 238000007499 fusion processing Methods 0.000 description 1
- 235000003642 hunger Nutrition 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000012423 maintenance Methods 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 229910052757 nitrogen Inorganic materials 0.000 description 1
- 230000003534 oscillatory effect Effects 0.000 description 1
- 230000003071 parasitic effect Effects 0.000 description 1
- 229910052698 phosphorus Inorganic materials 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 238000007789 sealing Methods 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
- 230000037351 starvation Effects 0.000 description 1
- 239000000126 substance Substances 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Classifications
-
- H—ELECTRICITY
- H01—ELECTRIC ELEMENTS
- H01M—PROCESSES OR MEANS, e.g. BATTERIES, FOR THE DIRECT CONVERSION OF CHEMICAL ENERGY INTO ELECTRICAL ENERGY
- H01M8/00—Fuel cells; Manufacture thereof
- H01M8/04—Auxiliary arrangements, e.g. for control of pressure or for circulation of fluids
- H01M8/04298—Processes for controlling fuel cells or fuel cell systems
- H01M8/04313—Processes for controlling fuel cells or fuel cell systems characterised by the detection or assessment of variables; characterised by the detection or assessment of failure or abnormal function
- H01M8/04664—Failure or abnormal function
- H01M8/04679—Failure or abnormal function of fuel cell stacks
-
- H—ELECTRICITY
- H01—ELECTRIC ELEMENTS
- H01M—PROCESSES OR MEANS, e.g. BATTERIES, FOR THE DIRECT CONVERSION OF CHEMICAL ENERGY INTO ELECTRICAL ENERGY
- H01M8/00—Fuel cells; Manufacture thereof
- H01M8/04—Auxiliary arrangements, e.g. for control of pressure or for circulation of fluids
- H01M8/04298—Processes for controlling fuel cells or fuel cell systems
- H01M8/04305—Modeling, demonstration models of fuel cells, e.g. for training purposes
-
- H—ELECTRICITY
- H01—ELECTRIC ELEMENTS
- H01M—PROCESSES OR MEANS, e.g. BATTERIES, FOR THE DIRECT CONVERSION OF CHEMICAL ENERGY INTO ELECTRICAL ENERGY
- H01M8/00—Fuel cells; Manufacture thereof
- H01M8/04—Auxiliary arrangements, e.g. for control of pressure or for circulation of fluids
- H01M8/04298—Processes for controlling fuel cells or fuel cell systems
- H01M8/04992—Processes for controlling fuel cells or fuel cell systems characterised by the implementation of mathematical or computational algorithms, e.g. feedback control loops, fuzzy logic, neural networks or artificial intelligence
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02E—REDUCTION OF GREENHOUSE GAS [GHG] EMISSIONS, RELATED TO ENERGY GENERATION, TRANSMISSION OR DISTRIBUTION
- Y02E60/00—Enabling technologies; Technologies with a potential or indirect contribution to GHG emissions mitigation
- Y02E60/30—Hydrogen technology
- Y02E60/50—Fuel cells
Landscapes
- Engineering & Computer Science (AREA)
- General Chemical & Material Sciences (AREA)
- Sustainable Development (AREA)
- Sustainable Energy (AREA)
- Manufacturing & Machinery (AREA)
- Chemical & Material Sciences (AREA)
- Chemical Kinetics & Catalysis (AREA)
- Electrochemistry (AREA)
- Life Sciences & Earth Sciences (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Automation & Control Theory (AREA)
- Artificial Intelligence (AREA)
- Computing Systems (AREA)
- Evolutionary Computation (AREA)
- Fuzzy Systems (AREA)
- Medical Informatics (AREA)
- Software Systems (AREA)
- Theoretical Computer Science (AREA)
- Health & Medical Sciences (AREA)
Abstract
The invention discloses a multi-fault diagnosis method based on an SOFC system, and belongs to the technical field of fuel cells. According to the invention, firstly, a fault model based on SOFC system air compressor faults, pile leakage and electrode layering is established, and state sensitivity analysis is carried out based on different faults so as to study the multi-fault diagnosis problem. And then, designing a state observer aiming at the temperature distribution and the gas mole fraction inside the electric pile which are greatly influenced by faults and are difficult to directly measure, and designing the temperature distribution and the gas mole fraction observer for the SOFC electric pile by adopting a supercoiled sliding mode algorithm and a one-dimensional electric pile model. Finally, in order to solve the composite fault diagnosis problem under the condition that multiple faults occur simultaneously, a SOFC fault decoupling diagnosis scheme based on a capsule network is provided. The composite characteristic is decoupled by using the capsule network, and the fault characteristic is represented by using a state with high sensitivity, so that the effect of multi-fault decoupling diagnosis can be achieved.
Description
Technical Field
The invention belongs to the technical field of fuel cells, and particularly relates to a multi-fault diagnosis method based on a Solid Oxide Fuel Cell (SOFC) system.
Background
The SOFC is a power generation device capable of directly converting chemical energy in fuel into electric energy under medium-high temperature conditions, has the characteristics of high power generation efficiency, low environmental pollution and the like, and is one of the most expected new energy sources at present. At present, the commercial application of the SOFC mainly faces the problems of high cost and short service life, and the fault diagnosis can timely discover and isolate the fault problem in the running of the system, thereby having great significance for improving the durability of the system and reducing the maintenance cost.
For SOFCs, the failure is not only a single occurrence, but is more common in the simultaneous occurrence of multiple failures. However, most of the existing fault diagnosis researches are concentrated on single fault diagnosis, and the diagnosis researches aiming at multiple faults of the system are deficient. When a system fails more, under the influence of coupling between faults, mixed characteristic situations can exist between different faults, and even a new characteristic can be presented, which makes fault diagnosis very challenging. In addition, due to the restriction of the high-temperature sealing condition of the SOFC, the internal temperature and the gas mole fraction of the electric pile cannot be directly measured, but are important state quantities affecting the system performance and reacting multiple faults, such as: pile leakage can affect the molar fraction of internal gas, cause local starvation and hot spot phenomena, and layering is mainly affected by temperature gradient. Thus, it is necessary to acquire the internal temperature and the gas mole fraction by means of state estimation to improve the accuracy of the multiple fault diagnosis.
In the existing multi-fault diagnosis, a fault diagnosis based on a model generally needs to establish an accurate fault model, and perform residual analysis and setting of a threshold value based on a predefined point. Data driven based fault tree diagnostic schemes can present indistinguishable situations in the face of mixed features. Although the deep learning based on sparse self-coding overcomes the problem of mixed characteristics, a large number of compound faults are needed for online characteristic learning, and the deep learning based on sparse self-coding is not easy to realize in actual engineering.
Disclosure of Invention
The invention provides a multi-fault diagnosis method based on an SOFC system, which is not only suitable for single fault diagnosis of the SOFC system, but also can carry out decoupling diagnosis when multiple faults occur simultaneously.
The invention adopts the technical scheme that:
a method of multiple fault diagnosis based on an SOFC system, the method comprising the steps of:
step 1, establishing a fault model of a solid oxide fuel cell SOFC system based on air compressor faults, stack leakage faults and electrode layering, and carrying out state sensitivity analysis based on the fault model to determine state quantity for fault diagnosis;
preferably, in the step, mathematical modeling is performed on SOFC peripheral equipment, a stack and faults based on electrochemical and thermodynamic principles, wherein the SOFC peripheral equipment adopts a modularized modeling mode, and the SOFC stack adopts a finite element modeling mode; carrying out process integration on the modeled model by adopting a module integration means, carrying out polarization characteristic curve verification and state sensitivity analysis on faults on the modeled model, and determining selected state quantity;
step 2, designing a state estimator aiming at the internal temperature and the gas mole fraction of the electric pile which have high fault sensitivity and are difficult to directly obtain in the selected state quantity so as to obtain a fault data set of the internal temperature and the gas mole fraction of the electric pile; the internal temperature and the gas mole fraction of the pile in the selected state quantity are subjected to data acquisition through a designed state estimator, and the rest state quantity can be directly obtained to acquire a corresponding fault data set;
preferably, the step 2 specifically comprises:
and (3) carrying out observer design on the temperature and gas distribution along the flow channel direction in the electric pile by adopting a state space equation according to an electric pile model, and obtaining a state equation form of the high-price sliding mode observer by additionally adding an output error affine term to a system state equation, wherein the state equation form is as follows:
wherein ,estimated value representing state +_>Representing status input +.>Representing the estimated output +.>Expressed as state->A related nonlinear equation; />Expressed as AND input u (stack inlet input, e.g. fuel input, air input) and status +.>A related nonlinear equation; />Then a disturbance-related nonlinear equation is expressed; />The affine injection matrix is expressed as an output error and is a decoupling matrix which is calculated by constructing a sliding mode observer at the time; />Representing a correction term; d is the load current disturbance (i.e. I load )，/>Representing a state estimation output equation.
Further, solving the decoupling matrix of the sliding mode observer according to the aboveTo reduce tremor during observation, the correction term +.>
Step 3, based on the acquired fault data set, diagnosing multiple faults of the SOFC system by adopting a fault decoupling diagnosis mode of a capsule network:
acquiring a fault data set corresponding to each state quantity determined in the step 1, and setting a corresponding fault type label for each sample data in the fault data set; performing data preprocessing on the sample data to construct input data of a characteristic extraction network based on a convolutional neural network;
inputting the sample data subjected to data preprocessing into a feature extraction network to perform feature extraction to obtain feature data of each sample data;
the method comprises the steps that feature data of sample data are subjected to dimension transformation to obtain input data of a capsule network classifier for fault classification, the capsule network classifier comprises a plurality of capsule layers, prediction probability of each fault type is output through a last capsule layer (the output of the last capsule layer is subjected to L2 norm to obtain the prediction probability of each fault type);
and performing deep learning training on the feature extraction network and the capsule network classifier based on a preset loss function, wherein when a preset training ending condition (reaching the maximum training times or the loss function value convergence requirement) is met, the multi-fault diagnosis device for the SOFC system is used for obtaining a fault diagnosis result, and when the fault diagnosis result is obtained, the fault type corresponding to the maximum prediction probability output by the last capsule layer of the capsule network classifier is the fault diagnosis result.
Further, the step 3 specifically includes the following steps:
step 301, preprocessing SOFC fault data
For the acquired fault data set, the fault data can be fused in a strapping mode (namely, a two-dimensional data matrix is formed through fusion processing so as to be matched with the dimension of input data of the network); after the data are subjected to bundling fusion, the data are subjected to mean normalization so as to be capable of carrying out feature extraction on SOFC fault data in a mode of processing pictures, and the corresponding mean normalization formula is as follows:
wherein Y is in the formula i An i-dimensional vector normalized for the mean;represented as a mean; delta represents the standard deviation.
Step 302, performing feature extraction on fault data by constructing a convolutional neural network;
after the SOFC failure data is preprocessed, the SOFC failure data can be described as shown in a formula (3)The matrix X of the dimension is used as the input of a convolution layer, the input of the convolution layer can be converted into the feature mapping shown in the formula (4), and the complete convolution layer output feature hidden emission can be obtained through the nonlinear activation function after the feature mapping obtained in the formula (4) is obtained as shown in the formula (5).
y i ＝ReLU(h i )＝max{0,h i } (5)
Wherein the symbols areRepresenting a convolution operation; ker (Ker) i Denoted as the ith convolution kernel (i=1, …, K); b i Bias terms for corresponding convolution kernels; x is x mj The number of slices for the sliding window; m corresponds to the characteristic dimension of fault diagnosis input, and n is the fault data length of each sample; j=1, …, D is expressed as the number of convolutional neurons passing through the convolutional layer, and D can be calculated according to formula (6); x (j) is the firstAn input matrix of j convolutional neurons; p is the number of Zero Padding (Zero Padding) at both input ends; s is denoted as the step size when the convolution kernel slides.
In equation (6), kernelSize represents the size of the convolution kernel and N represents the dimension of the convolved input data.
When the convolution layer outputs the mapping feature y i After that, there may be problems of excessively high information feature dimension and redundant features, so that the output of the previous layer needs to be subjected to the pooling layer to select the descending and main features, so as to avoid the overfitting behavior caused by noise and other unnecessary features under the condition of keeping the original information features as much as possible. The common Pooling layers mainly include Maximum Pooling (Maximum Pooling) and Average Pooling (Average Pooling), and the corresponding formulas are shown as follows:
wherein ,Wt Represented as a pooling window,represents the maximum pooled output of the kth layer,/->Represents the average pooling output of the k-th layer, h k,i Pooling window W representing a kth layer t The weight in the pool window is represented by a formula (7), wherein the corresponding weight in the pool window is selected and reserved for the feature mapping output in the upper layer; equation (8) is to sum and average all values under the pooling window to preserve.
When the SOFC fault data is subjected to feature extraction, the feature extraction is completed through the mutual alternation of multiple convolution layers and pooling layers, and the internal parameters can be automatically obtained according to the back propagation through a loss function in the training process, so that the number of convolution kernels of different layers is only required to be set, and the specific number of layers and the size of the convolution kernels for feature extraction can be set based on the actual scene requirement.
And 303, constructing a capsule decoupling network layer.
The fault data after being preprocessed and the feature extraction by the convolution module can be correspondingly described asIs transformed into N by dimension in D (D) in Vectors of dimension x P, i.e. N in Expressed as the number of input low-layer capsules, D in X P represents the vector dimension after feature extraction.
The input of the decoupled classifier may representThe calculation process from the low-layer capsule to the high-layer capsule between two adjacent layers is shown as the formula:
wherein ,representing the input of the ith primary capsule, W i,j Vector change matrix representing ith capsule and jth capsule, +.>Representing a predictive vector +.>Representing the coupling coefficient>Represents the mth capsule weighted output value, and squaring (·) represents the nonlinear activation function between capsules, ++>Output probability value representing nonlinear activation function, +.>Representing the weighted input of the primary capsule. />Is an output high-level capsule; n (N) out Expressed as the number of high-level capsules; d (D) out The X M is the dimension of a single high-layer capsule, and the number N of the finally decoupled layered capsules out The number of outputs is expressed as the number of types of faults, and the high-level capsule can pass through L after the dynamic routing protocol is calculated 2 Conversion of norms to probability values y for the respective classes pread ＝||.y HCaps || 2 。
Because of the need of decoupling the composite fault of the SOFC, the decision strategy of the decoupling classifier outputs the fault class corresponding to the fault label by setting a confidence threshold phi, namely:
wherein ,for class c predictive tag, +.>Expressed as category c present, +.>Representing the predictive probability output for category c.
When the capsule network performs decoupling classification on the composite fault of the SOFC, a boundary loss function is adopted as an optimization objective function, and the boundary loss function is as follows:
in the above, T c Is an indication function, which is expressed as T when the prediction sample belongs to the category c c The value of (1) is taken as 1, otherwise T c ＝0；m + When the fault category belongs to c, outputting a lower bound of a prediction probability value by the classifier; m is m - Outputting an upper bound of the prediction probability value by the classifier when the corresponding prediction fault category does not belong to the category c; λ is the weight penalty coefficient.
Further, in the fault diagnosis process, training is performed by using the normal and single fault data sets, and diagnostic testing is performed by using the single fault data set and the composite fault data.
The technical scheme provided by the invention has at least the following beneficial effects:
(1) The scheme provided by the invention not only can carry out fault diagnosis on single faults of the SOFC, but also can solve the characteristic problem of mixed faults and realize multi-fault decoupling diagnosis.
(2) When the multi-fault diagnosis scheme provided by the invention is used for carrying out multi-fault diagnosis on the SOFC, only a single fault data set is needed to carry out model training, and the problem that the composite fault of the SOFC actual known label is difficult to obtain can be solved.
(3) The embodiment of the invention is not only suitable for fault diagnosis of the pile node model, but also can be used for fault diagnosis of multiple piles in an expanding way. In addition, the present invention is applicable not only to the faults mentioned in this embodiment, but also to other types of fault diagnosis.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for the description of the embodiments will be briefly described below, and it is apparent that the drawings in the following description are only some embodiments of the present invention, and other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
FIG. 1 is a schematic diagram of a method of diagnosing SOFC system failure in accordance with an embodiment of the present invention;
fig. 2 is a basic structural diagram of an SOFC system of an embodiment of the present invention;
FIG. 3 is a block diagram of an implementation of a supercoiled sliding mode observer according to an embodiment of the invention;
FIG. 4 is a graph showing the verification of the polarization curve of a practical example of the present invention, wherein (4 a) is represented as a V-I characteristic response curve; (4 b) expressed as a P-I characteristic response curve;
FIG. 5 is a graph of estimation results according to an embodiment of the present invention, wherein (5 a) represents the temperature estimation results of the node 2; (5 b) represents the node 3 temperature estimation result; (5 c) represents the node 2 hydrogen mole fraction estimation result; (5 d) represents the node 3 hydrogen mole fraction estimation result;
FIG. 6 is a portion of an important fault data set referred to in an embodiment of the present invention, wherein (6 a) is denoted as air flow rate; (6 b) hydrogen flow rate; (6 c) is the node 3 stack temperature; (6 d) is the node 3 hydrogen mole fraction;
FIG. 7 is a schematic diagram of a data fusion approach to fault signatures in an embodiment of the present invention;
FIG. 8 is a graph of single failure diagnosis results for an SOFC system in accordance with an embodiment of the invention, wherein (8 a) is represented as a single failure test result; (8 b) a single fault confusion matrix map;
fig. 9 is a graph of diagnostic results of a composite failure of an SOFC system in an embodiment of the present invention.
Detailed Description
For the purpose of making the objects, technical solutions and advantages of the present invention more apparent, the embodiments of the present invention will be described in further detail with reference to the accompanying drawings.
The embodiment of the invention provides a multi-fault diagnosis method based on an SOFC system. Firstly, a fault model based on SOFC system air compressor faults, pile leakage and electrode layering is established, and state name and sensitivity analysis is carried out based on different faults so as to study the multi-fault diagnosis problem. And then, designing a state observer aiming at the temperature distribution and the gas mole fraction inside the electric pile which are greatly influenced by faults and are difficult to directly measure, and designing the temperature distribution and the gas mole fraction observer for the SOFC electric pile by adopting a supercoiled sliding mode algorithm and a one-dimensional electric pile model. Finally, in order to solve the composite fault diagnosis problem under the condition that multiple faults occur simultaneously, a SOFC fault decoupling diagnosis research scheme based on a capsule network is provided. The composite characteristic is decoupled by using the capsule network, and the fault characteristic is represented by using a state with high sensitivity, so that the effect of multi-fault decoupling diagnosis can be achieved.
As a possible implementation manner, as shown in fig. 1,2 and 3, the SOFC system-based multi-fault diagnosis method provided by the embodiment of the present invention includes the following steps:
step 1, establishing a one-dimensional model of the SOFC electric pile based on a finite element method:
referring to fig. 1 and 2, in the present embodiment, the SOFC stack is divided into 5 nodes, and modeling is performed with consideration of an electrochemical reaction equation, a gas transport equation, and a temperature equation for each node.
Wherein, the mathematical model of the electrochemical reaction equation is shown in formulas (15) - (18).
In the above-mentioned method, the step of,expressed as a pile output voltage value, i is the number of nodes and i ε {1, …,5}, ->Is open-circuit voltage, I is pile current, R ohm Is the ohmic value of the galvanic pile, R is the gas constant, F is the Faraday electromagnetic constant,/>For the temperature of the galvanic pile node, I L For limiting current, I 0 For the initial current +.>Nernst voltage for the ith node, < +.>Expressed as mole fractions of hydrogen, oxygen and water, respectively, for the ith node, a is the activation area and Le represents the electrolyte thickness.
Accordingly, the mathematical model of the gas transport equation is shown in equations (19) - (21). Wherein formula (19) describes a hydrogen transport process, formula (20) describes a water transport process, and formula (21) describes an oxygen transport process.
In the above-mentioned method, the step of, and />Respectively denoted as cathode inflow airDensity, air volume and anode fuel density and fuel volume; when i=1, +.>The stack inlet fuel and air flow rates are shown,stack fuel flow and air flow, respectively, representing the ith node, +.> and />The mole fractions of hydrogen, water and oxygen of the ith node and the mole fractions of hydrogen, water and oxygen of the inlet of the galvanic pile are respectively shown, and />The reaction rates of hydrogen, water, and oxygen, respectively, for the ith node can be described using equations (22) - (26).
In the above， and />Represents hydrogen, water and oxygen volumes, respectively, +.> and />The air flow rate and the fuel flow rate of the i-th node are respectively represented.
Correspondingly, the mathematical model of the temperature of the galvanic pile node is as follows:
in the above, ρ S 、V S Andrespectively representing the solid density, volume and specific heat capacity of the galvanic pile, +.> and />The average specific heat capacities of the anode fuel and the cathode gas, respectively expressed as the inlet fuel and air, and the ith node, ΔH R Indicating the reaction enthalpy. />Represented as voltage values at i nodes of the battery.
The I-V curve under the SOFC pile node model constructed according to the mathematical equation is shown in fig. 4, and the accuracy of the constructed model can be well verified from the I-V curve in the figure.
The air compressor fault model is as follows:
in the formula ,Tcp Indicating compressor outlet temperature, T 0 Representing inlet air temperature, beta is the air compression ratio of the compressor, eta is Is the isentropic efficiency of the compressor, k is the isentropic coefficient, c p,a Expressed as specific heat capacity, eta EM Expressed as motor efficiency, P cp Expressed as parasitic power, W air Expressed as peripheral air flow rate, delta expressed as the magnitude of the air compressor fault magnitude.
The pile leakage fault model corresponds to the following:
in the formula ,representing the mole fraction of the corresponding gas j of the ith node; />A constant value for the gas diffusion coefficient in the anode channels; />Expressed as anode-to-gas mole fraction; />Expressed as the cathode corresponds to the gas mole fraction, which can be calculated according to formulas (22) - (26).
The electrode layering fault model corresponds to the following:
in the formula ,and->Ohmic loss and activation loss for the ith node, A 0 For the initial electrochemical reaction area, κ and ζ are respectively represented as failure coefficients of 0-1, ++>Exchanging current density for the initial anode,/->Exchanging current density for the initial cathode, τ being the heat of reaction fraction; n is a reaction equivalent electron.
Step S201, designing a supercoiled sliding mode observer
When estimating the temperature distribution inside the galvanic pile, the internal 5 node temperatures are selected as state variables, so according to the temperature model and the gas transportation equation in the step 1, the internal 5 node temperatures can be described as the form of the state equation as follows:
in the formula ,a temperature state represented as an i-th node inside the stack; u (u) 1 ＝W f Represented as stack inlet fuel input, u 2 ＝W a Inlet air input, p, denoted as galvanic pile i The hydrogen stoichiometry, expressed as the i-th node of the stack; i is load disturbance current; />Voltage values represented as i nodes of the battery; c 1 ,c 2 ,…,c 12 The corresponding form can be expressed as shown in the following formula (35).
in the formula , and />Expressed as hydrogen specific heat capacity, oxygen specific heat capacity, water specific heat capacity and nitrogen specific heat capacity, respectively; />Mole fractions of hydrogen, oxygen, and water, respectively, expressed as the ith node; /> and />Respectively representing the mole fractions of the input hydrogen, oxygen and water; n represents the number of batteries ρ S 、V S and />Respectively representing the solid density, volume and specific heat capacity of the galvanic pile; and the anode channels and the cathode channels in the corresponding electric pile correspond to the mole fraction of the gasThe state space equations of the numbers can be described by equations (36) - (37).
wherein ,and->Is expressed as a constant calculated from the gas molar density and the gas volume inflow amount, and the distribution of the states of hydrogen and oxygen in the anode channels in the channels is described by the formula (36), and the distribution of the states of water in the cathode channels is described by the formula (37). />A calculation of the rate of hydrogen reaction, denoted as the ith node, which can be calculated by equations (22) - (26).
Here, in the present embodiment, by rewriting the expression (34) into the form of the observation equation of the expression (1), the observer equations corresponding to the temperature distribution and the gas mole fraction can be described as the following expressions (38) to (41).
wherein ,status value representing the i-th node, +.>Indicating the stack inlet temperature of (38) - (3)9) Is a temperature observer equation obtained according to formulas (34) - (35), and corresponding +.about.f. for the molar concentration of anode channel gas will be obtained according to formulas (36) - (37)> φ 2 (x) And cathode channel->φ 3 (x) The expressions are shown in the expression (40) and the expression (41), respectively.
Here, the observer designed in this embodiment is known from formulas (34) - (37)That is, only the easily measured outlet temperature of the electric pile and the mole fraction of the gas are needed to be known, so that an observer capable of estimating the internal temperature of the electric pile and the gas concentration distribution can be designed. Here, an affine injection matrix represented by the formula (1) is required>To ensure that the observer is stable and can track errors, the corresponding calculation formulas according to formulas (38) - (41) are as follows (36):
/>
in the formula , and />Denoted as ladderA degree operator; />Anddenoted as Li Daoshu and obtained according to formula (42)>The must-be matrix must be an observability matrix. And because B g ＝[0…1，0…1，0…1，0…，1] T Therefore, the corresponding temperature can be obtained>The matrix is shown in the following formula (43).
Here, in order to reduce the tremble caused by the observer while maintaining the robustness of the sliding mode observer, the error correction coefficient is performed by using the supercoiled sliding mode algorithm shown in formula (44)The specific update calculation procedure is shown in equations (44) - (45).
wherein ,the temperature estimation value of the galvanic pile node i is represented, the alpha and lambda parameters are convergence adjusting factors, different node devices can be provided with different values, the boundary condition corresponding to the delta sut function is 1, the simulation verification of the embodiment shows that the given value is alpha=0.2, lambda=0.6, and the used supercoiled water is adoptedThe algorithm is also not a strict sign function, which takes into account that the SOFC system itself will exhibit oscillatory behavior in time of fuel and air input.
In summary, the basic structure of the SOFC system in the embodiment of the present invention is shown in fig. 2: first, based on the constructed SOFC one-dimensional stack model (wherein,representing an estimate of the state input x, f () representing a state dependent nonlinear equation, g () representing a state dependent nonlinear equation and input to the stack inlet,/->Then represent a non-linear equation related to the disturbance, I load For load disturbance current, h () represents a state estimation output equation), under the current input parameters, a corresponding error term e of estimation output and measurement output can be obtained; and may be based on the error term e of each node i (subscript i indicates node device number) a corresponding error correction coefficient +.>α i 、λ i First and second convergence adjusting factors representing an ith node device, respectively; and then based on the error correction coefficient v of the current node equipment i Updating the estimated value (i.e. parameter epsilon corresponds to +.>To) to ultimately output a final estimated value, including:
temperature estimation value of galvanic pile nodeMolar fraction estimate of Hydrogen at galvanic pile node +.>Estimated value of mole fraction of water at galvanic pile node +.>Molar fraction estimate of oxygen at galvanic pile node +.>
S3.1, acquiring a fault data set.
As shown in fig. 1, the SOFC system fault model constructed by the embodiment of the present invention includes an air compressor fault, a stack leakage fault, and an electrode layering fault, and a part of the relevant acquired important fault data sets are shown in fig. 2. When data are collected, in order to ensure that the data of the system running under different working conditions can be obtained so as to avoid the accidental condition of a single working condition, the data sampling is carried out by selecting the fuel flow rate change range of [0.080 and 0.102] mol/s and the air flow rate change range of [0.710 and 0.858] mol/s.
After the normal data and three single fault data are obtained, the obtained fault data are subjected to data processing in a mean normalization mode, the corresponding mean normalization calculation is shown in a formula (2), the normalized data are subjected to data feature fusion in a mode shown in fig. 6 (namely, the 15-dimensional state features representing faults are uniformly bundled to form a whole to describe fault types), 15x 4000 data of each sample are obtained, 750 groups of samples of each type are obtained, and data represented by the data of 15x 4000 of each sample are shown in table 1.
TABLE 1
After preprocessing the normal data and the single fault data, dividing the normal data set and the single fault set into a training set and a testing set in a 2:1 mode, wherein the composite fault adopts the same preprocessing mode, only 250 groups are selected as the testing set and are not used for training, the data set adopts an One-hot coding mode, and the corresponding fault data and corresponding codes are shown in table 2.
TABLE 2
S302, convolutional network construction under feature extraction
When feature extraction is performed on a fault data set, feature extraction is performed in a form of three layers of convolution layers and two layers of average pooling layers, a convolution kernel of 15x16 is adopted in the first layer of convolution extraction, the convolution step length is the step length of (1 x 16), primary feature extraction is performed on convolution with the number of convolution filters being 32, and a Relu function is used for activation corresponding to an activation function, so that convolution output of (1,250,32) is obtained; then, a small kernel with the convolution kernel size of (1 x 3) is used for convolution, the corresponding convolution step length is 1x1, the number of convolution filters is 64, the zero complement is (0, 3), the feature mapping is carried out on the convolution layers, the Relu activation function is adopted for output, and the convolution output with the corresponding convolution step length of (1,254,64) is obtained; then, further extracting the characteristics by using an average pooling layer; and performing feature extraction by using a convolution layer with a convolution kernel dimension of (1 x 3), a convolution step length of 1x1, a filter number of 128 and zero-padding of (0, 2), wherein the corresponding obtained feature output is the feature output of (1,128,128), and finally performing feature extraction by using an average pooling layer.
S303, constructing a capsule diagnosis network.
After feature extraction is performed through a convolutional neural network to obtain (1,64,128) feature output, the feature output is transformed into 128 vectors with 64x1 dimension through dimension, and then the vectors with 64x1 dimension obtained after dimension transformation are input as low-level capsules, namely in the formula (9)The number of routes is 3, corresponding to the route transformation matrix +.>The capsule output activation function then employs the squaring function of equation (12), namely: the capsule output layer calculates according to formulas (9) - (12) to obtain high-grade capsule output with 64 (32, 64) outputs of the capsule layer 1; then the capsule layer 1 is transportedOut as input to the low-level capsule, selecting the routing transformation matrix +.>The routing times are selected to be 3, the output of the capsule layer 2 is calculated again according to the routing algorithm of the formulas (9) - (12), the output convolution module of the capsule layer 2 is obtained to be 32 (4,32) high-grade capsule outputs, and then the threshold phi=0.55 is selected for the output high-grade capsules, and the output high-grade capsules pass through L according to the formula (13) 2 The norm is converted into a probability value y of the corresponding output class (fault class) pread ＝||y HCaps || 2 。
In the embodiment of the present invention, the network parameters of the capsule diagnostic network used are shown in table 3:
TABLE 3 Table 3
Further, when the whole capsule network is optimally trained, training is performed by adopting normal data and a composite fault data set, and testing is performed by adopting the normal data, the single fault data and the composite fault data set.
Training the optimization network by correspondingly adopting the minimized boundary loss function of the (14), and when the predicted sample belongs to the category c, T c When the predicted samples do not belong to category c, t=1 c =0; lower bound m of classifier output prediction probability value + =0.9; the classifier outputs the upper bound m of the predictive probability value - =0.1; the weight penalty coefficient λ=0.2. Training is performed by using the normal data and the composite fault data set, and testing is performed by using the normal data, the single fault data and the composite fault data set.
The single fault diagnosis result is shown in fig. 8, the total test result of three faults and normal data is 98.8% from the prediction diagram of fig. 8a, the diagnosis model cannot be completely distinguished on the air compressor fault and the pile leakage fault from the confusion matrix of fig. 8b, but the prediction precision is 98% and the prediction precision of electrode layering faults is 99.2%.
The composite fault diagnosis result is shown in fig. 8, and the figure shows that the diagnosis model can completely separate the composite fault 6 (namely, the composite fault under the fault of pile leakage and the fault of electrode layering), and the total diagnosis precision of 4 composite faults can reach 96.243%, so that the diagnosis model can perform decoupling diagnosis on the composite fault under the fault of air compressor, the fault of pile leakage and the fault of electrode layering.
Finally, it should be noted that: the above embodiments are only for illustrating the technical solution of the present invention, and are not limiting; although the invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical scheme described in the foregoing embodiments can be modified or some technical features thereof can be replaced by equivalents; such modifications and substitutions do not depart from the spirit and scope of the technical solutions of the embodiments of the present invention.
What has been described above is merely some embodiments of the present invention. It will be apparent to those skilled in the art that various modifications and improvements can be made without departing from the spirit of the invention.
Claims (7)
1. A method for multiple fault diagnosis based on an SOFC system, comprising the steps of:
step 1, establishing a fault model of a solid oxide fuel cell SOFC system based on air compressor faults, stack leakage faults and electrode layering, and carrying out state sensitivity analysis based on the fault model to determine state quantity for fault diagnosis;
step 2, designing a state estimator for the internal temperature and the gas mole fraction of the electric pile in the state quantity determined in the step 1 so as to obtain a fault data set of the internal temperature and the gas mole fraction of the electric pile;
step 3, diagnosing multiple faults of the SOFC system by adopting a fault decoupling diagnosis mode of the capsule network:
step 301, obtaining a fault data set corresponding to each state quantity determined in step 1, and setting a corresponding fault type tag for each sample data in the fault data set; performing data preprocessing on the sample data to construct input data of a characteristic extraction network based on a convolutional neural network;
step 302, inputting the sample data after data preprocessing into a feature extraction network for feature extraction to obtain feature data of each sample data;
step 303, obtaining input data of a capsule network classifier for fault classification after dimension transformation of characteristic data of each sample data, wherein the capsule network classifier comprises a plurality of capsule layers, and outputting a prediction probability value of each fault type through a last capsule layer;
and performing deep learning training on the feature extraction network and the capsule network classifier based on a preset loss function, and when a preset training end condition is reached, obtaining a fault diagnosis result by the multi-fault diagnosis device for the SOFC system, wherein the fault type corresponding to the maximum prediction probability output by the last capsule layer of the capsule network classifier is the fault diagnosis result.
2. The method of claim 1, wherein in step 1, the SOFC peripheral device, the stack and the fault are mathematically modeled based on electrochemical and thermal principles, wherein the SOFC peripheral device is modeled in a modular manner and the SOFC stack is modeled in a finite element manner; and then adopting a module integration means to integrate the process of the modeled model, verifying the polarization characteristic curve of the modeled model, analyzing the state sensitivity of faults, and determining the selected state quantity.
3. The method according to claim 1, wherein in step 2, the state estimator is designed specifically by using the internal temperature of the electric pile and the gas mole fraction:
according to a pile model, adopting a state space equation to carry out observer design on temperature and gas distribution in the pile along the flow channel direction, and obtaining a state equation of a high-price sliding mode observer by additionally adding an output error affine term to a system state equation:
wherein ,representing state estimates>Representing status input +.>Representing the estimated output +.>Representation and status->Related nonlinear equation, +.>Representing AND input u and State->Related nonlinear equation, +.>Representing a nonlinear equation associated with the disturbance; />Affine injection matrix representing output errors, i.e. decoupling matrix,)>Representing estimated output->Correction terms from the actual measured output y, d representing load current disturbance, < >>Representing a state estimation output equation.
4. The method of claim 3, wherein the correction term is determined by a supercoiled sliding mode algorithm in the solving process of the state equation of the high-cost sliding mode observerAnd updating.
5. The method of claim 1, wherein in step 3, the feature extraction network is a network structure in which multiple convolution layers and pooling layers alternate with each other.
6. The method of claim 1, wherein in step 3, the feature extraction network and the capsule network classifier are deep learning trained using normal and single fault datasets, and diagnostic testing is performed using single fault datasets and composite fault datasets.
7. The method according to any one of claims 1 to 6, wherein a loss function is used as a boundary loss in deep learning training of the feature extraction network and the capsule network classifier, and the expression is:
wherein ,Tc Representing an indication function, T when the prediction sample belongs to category c c The value of (1) is taken as 1, otherwise T c The value of (2) is 0; m is m + When the fault category belongs to c, the capsule network classifier outputsOutputting a lower bound of the predicted probability value; m is m - When the predicted fault category does not belong to the category C, the capsule network classifier outputs the upper bound of the predicted probability value, C represents the fault category number, lambda is the weight penalty coefficient,representing the predictive probability output for category c.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310732313.5A CN116632295A (en) | 2023-06-20 | 2023-06-20 | Multi-fault diagnosis method based on SOFC system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310732313.5A CN116632295A (en) | 2023-06-20 | 2023-06-20 | Multi-fault diagnosis method based on SOFC system |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116632295A true CN116632295A (en) | 2023-08-22 |
Family
ID=87641918
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310732313.5A Pending CN116632295A (en) | 2023-06-20 | 2023-06-20 | Multi-fault diagnosis method based on SOFC system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116632295A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117874665A (en) * | 2024-03-13 | 2024-04-12 | 西北工业大学宁波研究院 | SOFC system multi-fault diagnosis method and system |
-
2023
- 2023-06-20 CN CN202310732313.5A patent/CN116632295A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117874665A (en) * | 2024-03-13 | 2024-04-12 | 西北工业大学宁波研究院 | SOFC system multi-fault diagnosis method and system |
| CN117874665B (en) * | 2024-03-13 | 2024-05-10 | 西北工业大学宁波研究院 | SOFC system multi-fault diagnosis method and system |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Zhou et al. | Degradation prediction of PEM fuel cell stack based on multiphysical aging model with particle filter approach | |
| Rotondo et al. | Robust fault diagnosis of proton exchange membrane fuel cells using a Takagi-Sugeno interval observer approach | |
| CN110579709B (en) | Fault diagnosis method for proton exchange membrane fuel cell for tramcar | |
| CN107180983B (en) | Fault diagnosis method and system for solid oxide fuel cell stack | |
| Kim et al. | State-of-health diagnosis based on hamming neural network using output voltage pattern recognition for a PEM fuel cell | |
| Yang et al. | Modeling and optimization for proton exchange membrane fuel cell stack using aging and challenging P systems based optimization algorithm | |
| CN103336867B (en) | Proton Exchange Membrane Fuel Cells model optimization disposal route | |
| Wu et al. | A hybrid prognostic model applied to SOFC prognostics | |
| CN116632295A (en) | Multi-fault diagnosis method based on SOFC system | |
| CN109840593B (en) | Method and apparatus for diagnosing solid oxide fuel cell system failure | |
| CN114899457B (en) | Fault detection method for proton exchange membrane fuel cell system | |
| Guarnieri et al. | A selective hybrid stochastic strategy for fuel-cell multi-parameter identification | |
| Peng et al. | Generalized spatial–temporal fault location method for solid oxide fuel cells using LSTM and causal inference | |
| Fathy et al. | Recent approach based heterogeneous comprehensive learning Archimedes optimization algorithm for identifying the optimal parameters of different fuel cells | |
| CN116231000A (en) | Method for constructing fuel cell/electric pile simulation model | |
| Nagulapati et al. | Machine learning based fault detection and state of health estimation of proton exchange membrane fuel cells | |
| CN115598536A (en) | PEMFC fault diagnosis method based on fuzzy C-means clustering and probabilistic neural network | |
| He et al. | A quick evaluation method for the lifetime of the fuel cell MEA with the particle filter algorithm | |
| CN104778315B (en) | Merge the fuel cell Optimization Modeling method of krill behavior artificial bee colony algorithm | |
| Sun et al. | Fault diagnosis method for proton exchange membrane fuel cell system based on digital twin and unsupervised domain adaptive learning | |
| CN115993551A (en) | Lithium ion battery diagnosis method integrating physical information parameters and electrochemical impedance spectrum | |
| Xu et al. | Self-adaptive henry gas solubility optimizer for identification of solid oxide fuel cell | |
| CN114976128A (en) | PEMFC fault prediction and health management system and method based on five-dimensional digital twin technology | |
| Zhang | Contribution to prognostics of proton exchange membrane fuel cells: approaches based on degradation information at multiple levels | |
| Abbou et al. | A novel approach for predicting PEMFC in varying ambient conditions by using a transient search optimization algorithm based on a semiempirical model. |
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 |