CN116187195A - Fault diagnosis method and system for proton exchange membrane fuel cell integrated system - Google Patents

Fault diagnosis method and system for proton exchange membrane fuel cell integrated system Download PDF

Info

Publication number
CN116187195A
CN116187195A CN202310218121.2A CN202310218121A CN116187195A CN 116187195 A CN116187195 A CN 116187195A CN 202310218121 A CN202310218121 A CN 202310218121A CN 116187195 A CN116187195 A CN 116187195A
Authority
CN
China
Prior art keywords
model
integrated system
pemfc
fault
fuel cell
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
Application number
CN202310218121.2A
Other languages
Chinese (zh)
Inventor
赵波
陈哲
章雷其
吴启亮
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
State Grid Zhejiang Electric Power Co Ltd
Electric Power Research Institute of State Grid Zhejiang Electric Power Co Ltd
Original Assignee
State Grid Zhejiang Electric Power Co Ltd
Electric Power Research Institute of State Grid Zhejiang Electric Power Co Ltd
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by State Grid Zhejiang Electric Power Co Ltd, Electric Power Research Institute of State Grid Zhejiang Electric Power Co Ltd filed Critical State Grid Zhejiang Electric Power Co Ltd
Priority to CN202310218121.2A priority Critical patent/CN116187195A/en
Publication of CN116187195A publication Critical patent/CN116187195A/en
Pending legal-status Critical Current

Links

Images

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
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
    • G06F17/10—Complex mathematical operations
    • G06F17/11—Complex mathematical operations for solving equations, e.g. nonlinear equations, general mathematical optimization problems
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/20—Design optimisation, verification or simulation
    • G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
    • 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
    • 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
    • H01M8/04313—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
    • H01M8/04686—Failure or abnormal function of auxiliary devices, e.g. batteries, capacitors
    • 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
    • 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
    • H01M8/10—Fuel cells with solid electrolytes
    • H01M8/1016—Fuel cells with solid electrolytes characterised by the electrolyte material
    • H01M8/1018—Polymeric electrolyte materials
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2119/00—Details relating to the type or aim of the analysis or the optimisation
    • G06F2119/02—Reliability analysis or reliability optimisation; Failure analysis, e.g. worst case scenario performance, failure mode and effects analysis [FMEA]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2119/00—Details relating to the type or aim of the analysis or the optimisation
    • G06F2119/04—Ageing analysis or optimisation against ageing
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2119/00—Details relating to the type or aim of the analysis or the optimisation
    • G06F2119/08—Thermal analysis or thermal optimisation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2119/00—Details relating to the type or aim of the analysis or the optimisation
    • G06F2119/14—Force analysis or force optimisation, e.g. static or dynamic forces
    • 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
    • H01M8/10—Fuel cells with solid electrolytes
    • H01M2008/1095—Fuel cells with polymeric electrolytes
    • 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)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • General Chemical & Material Sciences (AREA)
  • General Physics & Mathematics (AREA)
  • Chemical Kinetics & Catalysis (AREA)
  • Electrochemistry (AREA)
  • Sustainable Energy (AREA)
  • Sustainable Development (AREA)
  • Manufacturing & Machinery (AREA)
  • Chemical & Material Sciences (AREA)
  • Mathematical Physics (AREA)
  • Evolutionary Computation (AREA)
  • Software Systems (AREA)
  • Data Mining & Analysis (AREA)
  • General Engineering & Computer Science (AREA)
  • Artificial Intelligence (AREA)
  • Health & Medical Sciences (AREA)
  • Pure & Applied Mathematics (AREA)
  • Computing Systems (AREA)
  • Mathematical Optimization (AREA)
  • Medical Informatics (AREA)
  • Mathematical Analysis (AREA)
  • Computational Mathematics (AREA)
  • Automation & Control Theory (AREA)
  • Fuzzy Systems (AREA)
  • Computer Hardware Design (AREA)
  • Operations Research (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Algebra (AREA)
  • Geometry (AREA)
  • Databases & Information Systems (AREA)
  • Computational Linguistics (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Fuel Cell (AREA)

Abstract

The invention discloses a fault diagnosis method and system for a proton exchange membrane fuel cell integrated system. The fault diagnosis method adopts the following technical scheme: establishing a PEMFC integrated system model according to an electrochemical reaction mechanism and an empirical formula in the dynamic working condition operation process of the fuel cell; performing verification analysis on the constructed mathematical model of the PEMFC integrated system by utilizing experimental data, simulating output characteristic data of the PEMFC integrated system in different running states, and storing fault characteristic images; extracting and storing output characteristic data of the pre-training model into GoogLeNet by adopting transfer learning, modifying parameters of a full link layer and an output layer in the GoogLeNet, and training fault characteristic images; and finally, classifying the fault characteristic images by using the trained GoogLeNet model. The invention can effectively diagnose five running states of the PEMFC integrated system, such as normal, cooling system fault, hydrogen supply system fault, air supply fault and flooding fault.

Description

Fault diagnosis method and system for proton exchange membrane fuel cell integrated system
Technical Field
The invention relates to the technical field of hydrogen energy, in particular to a fault diagnosis method and system for a Proton Exchange Membrane Fuel Cell (PEMFC) integrated system based on a GoogLeNet convolutional neural network and transfer learning.
Background
The fault diagnosis method of Proton Exchange Membrane Fuel Cells (PEMFC) is mainly divided into two major categories, model-based and non-model-based.
The model-based fault diagnosis method generates a residual by comparing an actual measured value of a system with an output value of an established model, and then completes fault diagnosis through residual analysis and decision. However, for such complex proton exchange membrane fuel cell systems, which are coupled with multiple physical fields involving electrochemistry, electromagnetics, thermodynamics, and fluid dynamics, achieving accurate model-based PEMFC fault diagnosis presents a significant challenge.
The non-model-based PEMFC fault diagnosis method mainly comprises an experimental test-based method and a data driving-based method. The method based on the experimental test mainly detects and diagnoses the abnormal condition of the PEMFC through the electromagnetic and fault operation and other experimental tests, and has the advantages of no need of collecting a large amount of data and small invasiveness to the fuel cell. However, the method based on the experimental test requires a high-precision experimental instrument, and the expensive equipment cost is a problem which cannot be ignored. The diagnosis method based on data driving does not need to establish a complex mechanism model, and the diagnosis process does not involve the operation of related experimental instruments, and the main idea is to consider the diagnosis process as a black box, and the machine learning algorithm is adopted to diagnose the collected historical fault data. Neural networks are widely used as a powerful and effective fault diagnosis tool to solve highly complex nonlinear pattern recognition or classification problems. The neural network is a core algorithm in the field of image recognition, has great advantages in the aspect of image processing, but the current research is based on vector data for fault diagnosis, and the current research has not been performed for judging the working state of the fuel cell system by utilizing the state image characteristics from the angles of dynamic operation conditions such as start-stop, idle speed, load variation and the like.
Disclosure of Invention
In order to solve the defects in the prior art, the invention provides a fault diagnosis method and a fault diagnosis system for a PEMFC integrated system based on a GoogleNet convolutional neural network and transfer learning, which are used for combining the GoogleNet convolutional neural network image identification method and the transfer learning to be applied to the fault diagnosis of a proton exchange membrane fuel cell for the first time.
In order to achieve the above purpose, the invention adopts a technical scheme as follows: a method for diagnosing faults of an integrated system of a proton exchange membrane fuel cell, comprising the steps of:
1) According to an electrochemical reaction mechanism of the proton exchange membrane fuel cell and an empirical model formula under a dynamic working condition, constructing a mathematical model of a PEMFC integrated system, wherein an auxiliary machine system comprises a cooling system, an air supply system and a hydrogen supply system;
2) Verifying a built mathematical model of the PEMFC integrated system according to experimental data, and analyzing the performance and power consumption of the PEMFC integrated system under a dynamic working condition;
3) Obtaining output characteristic data of five states of normal, cooling system faults, hydrogen supply system faults, air supply system faults and flooding faults by changing parameters of each component of the PEMFC integrated system;
4) Selecting some output characteristic data of the PEMFC integrated system as fault diagnosis characteristic quantity and outputting characteristic images of the selected fault diagnosis characteristic quantity;
5) Performing weight migration on the pre-trained Googlene model by adopting migration learning to obtain an optimized Googlene convolutional neural network classification model; and finally, diagnosing the obtained characteristic images in five states by utilizing the classification model, and analyzing the result.
Firstly, according to the reaction mechanism of the fuel cell stack under the actual operation dynamic working conditions such as start-up and stop, idling, load change and the like, establishing a PEMFC integrated system mathematical model comprising a fuel cell, a cooling system, an air supply system and a hydrogen supply system. And secondly, verifying and analyzing the constructed mathematical model of the PEMFC integrated system by utilizing experimental data, simulating the output characteristic data of the PEMFC integrated system in different running states, and storing fault characteristic images. And then, extracting and storing the output characteristic data part of the pre-training model into GoogLeNet by adopting transfer learning, modifying parameters of a full link layer and an output layer and training fault characteristic images. And finally, classifying the fault characteristic images by using the trained GoogLeNet model.
In the step 3), the output characteristic data includes a plurality of data such as current, voltage, electrical efficiency, thermal efficiency, temperature, cathode and anode intake pressure.
Further, in the step 1), the PEMFC integration system mathematical model includes:
1) Fuel cell output characteristic model
The cell output voltage of the stack is expressed by the following formula:
V cell ＝E Nernst -V act -V ohm -V conc (2)
wherein V is cell Is the output voltage of the single cell; e (E) Nernst Is a thermodynamic electromotive force; v (V) act For activating polarization overvoltage; v (V) ohm Is an ohmic polarization overvoltage; v (V) conc Is concentration polarization overvoltage;
pile output voltage V stack Expressed as:
V stack ＝n·V cell (7)
wherein n is the number of cells constituting the stack;
electric energy P generated by PEMFC integrated system el_st And electric power P th_st The method comprises the following steps of:
P el_st ＝n·V cell ·i·A (8)
P th_st ＝n·(V L -V stack )·i (9)
wherein V is L I is the operating current of the electric pile, A is the surface area of the proton exchange membrane;
the electrical efficiency of the stack is calculated from the following formula:
Figure BDA0004115691840000031
wherein mu is f For the fuel utilization coefficient of the fuel to be used,
Figure BDA0004115691840000032
is the chemical coefficient of hydrogen;
2) Cooling system
The energy consumption of the cooling system is calculated as follows:
Figure BDA0004115691840000033
in xi 5 ，ξ 6 ，ξ 7 And xi 8 Is an empirical constant; f (f) w For cooling water flow, the calculation formula is as follows:
Figure BDA0004115691840000034
in xi 9 Is an empirical constant; ρ w Is the density of water;
Figure BDA0004115691840000035
is the specific heat capacity of water under normal conditions; delta T w For the temperature difference between the cooling water entering and leaving the PEMFC; p (P) humid For gas humidification energy consumption, it approximates the energy consumption of inlet temperature water into saturated steam:
Figure BDA0004115691840000036
in the method, in the process of the invention,
Figure BDA0004115691840000037
and->
Figure BDA0004115691840000038
The flow rates of the water vapor contained in the humidified air and hydrogen are respectively; />
Figure BDA0004115691840000039
Enthalpy consumed for heating 1 mole of water to water vapor at an inlet temperature of 298.15K:
Figure BDA00041156918400000310
in the method, in the process of the invention,
Figure BDA00041156918400000311
is the water molar mass under normal conditions; t (T) stack Is the temperature of the galvanic pile;
3) Air supply system
The energy consumption of the air supply system in the air intake pressurization is calculated as follows:
P comp ＝c p ·ΔT gas ·Q air (15)
wherein, c p Constant pressure specific heat capacity for inlet gas; q (Q) air Air intake mass flow; delta T gas For the temperature rise of the compressor inlet gas, the calculation method is as follows:
Figure BDA00041156918400000312
wherein p is 1 Is the pressure before the gas is compressed; p is p 2 Is the pressure of the compressed gas; gamma is the ratio of the specific heat capacity of the gas to the specific heat capacity of the constant pressure; t (T) 2 T is the temperature of the gas before being introduced into the compressor 1 The temperature of the gas after being introduced into the compressor;
4) Hydrogen supply system
The hydrogen circulation pump recovers unreacted hydrogen fuel, and the calculation result of the ideal gas adiabatic compression process is regarded as the power consumption of the circulation pump:
Figure BDA0004115691840000041
in which W is cyc Is the mass flow of the circulation pump;
Figure BDA0004115691840000042
is the specific heat capacity of hydrogen at constant pressure; t (T) t Is the gas precompression temperature; pi cyc Is an adiabatic compression ratio; η (eta) cyc Efficiency as a hydrogen circulation pump;
system electric power P ele_sys Electric efficiency eta ele_sys And system thermal efficiency eta th_sys The calculation method of (2) is as follows:
Figure BDA0004115691840000043
further, in the step 2), the performance and power consumption of the proton exchange membrane fuel cell integrated system include a stack operating temperature, a polarization curve, thermal efficiency and electric power of the proton exchange membrane fuel cell integrated system.
Further, in the step 3), the failure generation mechanism is as shown in table 1:
TABLE 1 failure generation mechanism
Fault type Mechanism of production
Cooling system failure (F1) Shortening coolant channel, changing laminar flow and turbulent flow Reynolds number of radiator
Hydrogen supply system failure (F2) Reducing hydrogen excess ratio and intake pressure
Air supply system failure (F3) Reducing air excess ratio and intake pressure
Flooding failure (F4) Reducing intake relative humidity
Further, in the step 4), the selected fault diagnosis feature values are shown in table 2:
TABLE 2 fault diagnosis feature
Figure BDA0004115691840000044
Figure BDA0004115691840000051
Further, in the step 5), the main structure of google net is composed of 9 association modules; the association module assembles 3 convolution kernels with different sizes and a pooling operation into a network module, and the whole network structure is assembled by taking the network module as a unit when designing the neural network; the association module processes the input images in four different modes, and after splicing and fusing in characteristic dimension, the input images are transmitted to the next-layer association module;
each faulty sample is first processed by a convolution operation in the hidden layer, the output of which is defined as follows:
Figure BDA0004115691840000052
wherein f represents a nonlinear activation function; * Is a convolution operator; n is the number of convolution kernels; omega j And b j Respectively representing the weight and the bias value; v i Is the ith feature image;
pooling operation merges the n×1 faulty image block outputs of the previous convolutional layer into a single value of the next layer, thereby reducing the size of the feature vector; each pooling layer follows the previous convolutional layer, learns the edges and texture of the input fault state image using a max pooling operation, the output of which is calculated as follows:
Figure BDA0004115691840000053
wherein u (n, 1) is a window function of the previous layer of fault image block; a, a j Is the maximum value in the image block;
the local features extracted through rolling and pooling operations are recombined through a full-connection layer, and the full-connection layer is connected with an output layer, wherein the output layer consists of two neurons representing landslide and non-landslide; all parameters of the CNN layer are optimized using a back-propagation algorithm, the purpose of which is to minimize the loss value calculated from the loss function, defined as follows:
Figure BDA0004115691840000054
wherein m is the number of input landslide data, y i And
Figure BDA0004115691840000055
the true label and the predicted label of the ith input fault sample are represented respectively.
Still further, in step 5), a google net convolutional neural network is used as a pre-training model for the migration learning, which is trained from a subset of the large database of imagenets, and although imagenets do not contain ultrasound images, the original features of the spots, colors and lines extracted by the bottom convolution are present in almost all images, and by means of the migration learning, these original features are identified by the pre-training network and replaced with the last layers for combining features and completing the final classification.
In step 5), fine tuning of the target network model is required to reduce redundant parameters in the original model, improve generalization capability of the fault diagnosis model, and save training time.
Still further, in the step 5), the fine tuning step is as follows: firstly, transferring parameters of the first three associated modules of the pre-training model to a target model and freezing; and then, the initial weights of the parameters of the six associated modules after the pre-training model are reused as initial training values of the target model, and the parameters are updated continuously in the training process.
The invention adopts another technical scheme that: a proton exchange membrane fuel cell integrated system fault diagnosis system, comprising:
mathematical model building unit: according to an electrochemical reaction mechanism of the proton exchange membrane fuel cell and an empirical model formula under a dynamic working condition, constructing a mathematical model of a PEMFC integrated system, wherein an auxiliary machine system comprises a cooling system, an air supply system and a hydrogen supply system;
mathematical model verification unit: verifying a built mathematical model of the PEMFC integrated system according to experimental data, and analyzing the performance and power consumption of the PEMFC integrated system under a dynamic working condition;
an output characteristic data acquisition unit: obtaining output characteristic data of five states of normal, cooling system faults, hydrogen supply system faults, air supply system faults and flooding faults by changing parameters of each component of the PEMFC integrated system;
a feature image output unit: selecting some output characteristic data of the PEMFC integrated system as fault diagnosis characteristic quantity and outputting characteristic images of the selected fault diagnosis characteristic quantity;
a feature image diagnosis unit: performing weight migration on the pre-trained Googlene model by adopting migration learning to obtain an optimized Googlene convolutional neural network classification model; and finally, diagnosing the obtained characteristic images in five states by utilizing the classification model, and analyzing the result.
The invention has the following beneficial effects: establishing a PEMFC integrated system model according to an electrochemical reaction mechanism and an empirical formula in the dynamic working condition operation process of the fuel cell; then, verifying the model by adopting experimental data, and changing parameters of a model part to generate a characteristic fault image data set; and finally, migrating the weight in the pre-training model to a GoogLeNet model by adopting migration learning, improving the convergence rate and generalization capability of the classification model, and effectively judging five running states of normal PEMFC integrated systems, cooling system faults, hydrogen starvation, air starvation and flooding faults.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following description of the embodiments or the drawings used in the description of the prior art will make a brief description; it will be apparent to those of ordinary skill in the art that the drawings in the following description are of some embodiments of the invention and that other drawings may be derived from them without undue effort.
FIG. 1 is a block diagram of a PEMFC integrated system of the present invention;
FIG. 2 is a diagram of a deep learning model based on GoogLeNet of the present invention;
FIG. 3 is a flow chart of the transfer learning of the present invention;
fig. 4 is a graph showing output characteristics of a fuel cell stack in various states of the present invention.
Detailed Description
For the purpose of making the objects, technical solutions and advantages of the embodiments of the present invention more apparent, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention, and it is apparent that the described embodiments are some embodiments of the present invention, but not all embodiments of the present invention. All other embodiments, which can be made by one of ordinary skill in the art without inventive faculty, are intended to be within the scope of the present invention, based on the embodiments of the present invention.
Example 1
The embodiment is a fault diagnosis method of a PEMFC integrated system based on a GoogLeNet convolutional neural network and transfer learning, comprising the following steps:
1) According to the electrochemical reaction mechanism of the fuel cell and an empirical model formula under a dynamic working condition, a mathematical model of a PEMFC integrated system is built, wherein an auxiliary machine model comprises a cooling system model, an air supply system model and a hydrogen supply system model;
2) Verifying a built mathematical model of the PEMFC integrated system according to experimental data, and analyzing the performance and power consumption of the system under a dynamic working condition;
3) Obtaining output characteristic data of five states of a normal state, a cooling system fault, a hydrogen supply system fault (also called a hydrogen starvation fault), an air supply system fault (also called an air starvation fault) and a flooding fault by changing parameters of all components of the PEMFC integrated system;
4) Selecting some output characteristic data of the PEMFC integrated system as fault diagnosis characteristic quantity and outputting characteristic images of the selected fault diagnosis characteristic quantity;
5) Performing weight migration on the pre-trained Googlene model by adopting migration learning to obtain an optimized Googlene convolutional neural network classification model; and finally, diagnosing the characteristic image data of the five states by using the model, and analyzing the result.
In the step 1), the structure of the PEMFC integrated system is shown in fig. 1, and the mathematical model is as follows:
fuel cell output characteristic model:
in order to clearly analyze the failure of the PEMFC system, it is first necessary to study the PEMFC dynamic model. PEMFC is a low temperature fuel cell whose operating temperature is 55 ℃, and the reaction product is liquid water. The total reaction in PEMFC is expressed as follows:
Figure BDA0004115691840000071
the output voltage of a single cell can be expressed by the following formula:
V cell ＝E Nernst -V act -V ohm -V conc (2)
wherein V is cell Is the output voltage of the single cell; e (E) Nernst Is a thermodynamic electromotive force; v (V) act For activating polarization overvoltage; v (V) ohm Is an ohmic polarization overvoltage; v (V) conc Is concentration polarization overvoltage.
The thermodynamic electromotive force is calculated as:
Figure BDA0004115691840000081
/>
wherein Δg is the free energy change; f is Faraday constant; Δs is the variation of entropy; t (T) stack Is the temperature of the galvanic pile; r is R u Is a gas constant;
Figure BDA0004115691840000082
and->
Figure BDA0004115691840000083
The partial pressures of hydrogen and oxygen at the cathode and anode are respectively.
The calculation formula of the activation polarization overvoltage is as follows:
Figure BDA0004115691840000084
in xi 1 ，ξ 2 ,，ξ 3 And xi 4 As a rule of thumb constant,
Figure BDA0004115691840000085
for cathode oxygen concentration, i is stack operating current.
Ohmic polarization overvoltage V ohm The method is calculated according to the following formula:
Figure BDA0004115691840000086
wherein R is ohm Is ohmic resistance, r m The specific resistance of the proton exchange membrane is that of l, the thickness of the proton exchange membrane is that of A, and the surface area of the proton exchange membrane is that of A.
Concentration polarization overvoltage V conc The method is calculated according to the following formula:
Figure BDA0004115691840000087
wherein i is max For maximum current density through the stack.
Pile output voltage V stack Can be expressed as:
V stack ＝n·V cell (7)
where n is the number of cells that make up the PEMFC stack.
The electric energy and the electric power generated by the PEMFC system are respectively as follows:
P el_st ＝n·V cell ·i·A (8)
P th_st ＝n·(V L -V stac k)·i (9)
wherein V is L Is equivalent voltage at low hydrogen heating value.
The electrical and thermal efficiency of the stack is calculated from the following formula:
Figure BDA0004115691840000091
wherein mu is f For the fuel utilization coefficient of the fuel to be used,
Figure BDA0004115691840000092
is the chemical coefficient of hydrogen.
Cooling system model:
the auxiliary components of the cooling system consume a large amount of power and have a great influence on the overall system performance, and their energy consumption is calculated as follows:
Figure BDA0004115691840000093
in xi 5 ，ξ 6 ，ξ 7 And xi 8 Is an empirical constant, f w For cooling water flow, the calculation formula is as follows:
Figure BDA0004115691840000094
in xi 9 Is an empirical constant; ρ w Is the density of water;
Figure BDA0004115691840000095
is the specific heat capacity of water under normal conditions; delta T w For the temperature difference between the cooling water entering and leaving the PEMFC; p (P) humid For gas humidification energy consumption, it approximates the energy consumption of inlet temperature water into saturated steam: />
Figure BDA0004115691840000096
In the method, in the process of the invention,
Figure BDA0004115691840000097
and->
Figure BDA0004115691840000098
The flow rates of the water vapor contained in the humidified air and hydrogen are respectively; />
Figure BDA0004115691840000099
Enthalpy consumed for heating 1 mole of water to water vapor at an inlet temperature of 298.15K:
Figure BDA00041156918400000910
in the method, in the process of the invention,
Figure BDA00041156918400000911
is the molar mass of water under normal conditions.
Model of air supply system:
an air compressor is generally used to pressurize the inlet air to improve the operation of the cathode side of the PEMFC. The energy consumption of the air supply system in the air intake pressurization is calculated as follows:
P comp ＝c p ·ΔT gas ·Q air (15)
wherein, c p Constant pressure specific heat capacity for inlet gas; q (Q) air Air intake mass flow; delta T gas For the temperature rise of the compressor inlet gas, the calculation method is as follows:
Figure BDA00041156918400000912
wherein p is 1 Is the pressure before the gas is compressed; p is p 2 Is the pressure of the compressed gas; gamma is the ratio of the specific heat capacity of the gas to the specific heat capacity of the constant pressure.
Hydrogen supply system model:
the hydrogen circulation pump can recover unreacted hydrogen fuel, and the calculation result of the ideal adiabatic compression process is regarded as the power consumption of the circulation pump:
Figure BDA0004115691840000101
in which W is cyc Is the mass flow of the circulation pump;
Figure BDA0004115691840000102
is the specific heat capacity of hydrogen at constant pressure; t (T) t Is the gas precompression temperature; pi cyc Is an adiabatic compression ratio; η (eta) cyc Is the efficiency of the hydrogen circulation pump.
System electric power P ele_sys Electric efficiency eta ele_sys And system thermal efficiency eta th_sys The calculation method of (2) is as follows:
Figure BDA0004115691840000103
in the step 2), the performance and the power consumption comprise the operating temperature, polarization curve, thermal efficiency, electric power and the like of a pile of the proton exchange membrane fuel cell integrated system.
In the step 3), the fault generating method is as shown in table 1:
TABLE 1 failure generation mechanism
Fault type Mechanism of production
Cooling system failure (F1) Shortening coolant channel, changing laminar flow and turbulent flow Reynolds number of radiator
Starvation of Hydrogen (F2) Reducing hydrogen excess ratio and intake pressure
Air hunger (F3) Reducing air excess ratio and intake pressure
Flooding failure (F4) Reducing intake relative humidity
As shown in fig. 4, the output characteristics of the fuel cell in different failure states may be significantly different. When the cooling system fails, heat generated by the PEMFC integrated system cannot be discharged in time, so that the operation temperature of the electric pile is abnormal, and the balance and stability of the operation of the electric pile are destroyed. The water content of the proton exchange membrane can be reduced at high temperature, and compared with a stable working condition, the local heating caused by the abnormal operation at high temperature under a dynamic working condition is more remarkable, and the local temperature of the fuel cell stack is too high under a severe condition, so that the membrane is broken. Meanwhile, the output voltage of the PEMFC integrated system gradually decreases along with the increase of current, and the concentration polarization overvoltage and the ohmic polarization overvoltage are in direct proportion to the current, so that the output voltage of the electric pile decreases rapidly under high current density. Cooling system failure can exacerbate thermal imbalance inside the stack, resulting in reduced thermal efficiency and electrical power of the PEMFC system.
When the air starvation fault occurs, the local concentrated reaction of the inlet area of the PEMFC system is enhanced, so that a large temperature gradient exists on the proton exchange membrane, and the water content of the proton exchange membrane is far lower than the standard state. The PEMFC integrated system is in an air starvation fault and dehydration state under a dynamic working condition for a long time, and can cause membrane dryness and even membrane degradation, which can cause irreversible damage to the PEMFC integrated system. Therefore, the air starvation fault may cause the PEMFC system to significantly degrade, and the output voltage, the thermal conversion efficiency, and the electric power may be affected to some extent.
When the hydrogen starvation failure occurs, a reverse potential failure may occur, which causes the anode catalyst layer to degrade and leave holes in the proton exchange membrane due to local heat generation under the driving of the adjacent cells. Damage to components will greatly affect the output performance of the PEMFC integrated system, and the continuously variable operation will also accelerate the destruction of the heat balance inside the fuel cell stack, so that the output voltage, heat efficiency and electric power of the PEMFC integrated system are lower than those in a normal state.
Water management balance in PEMFC integrated systems is critical, which affects the stability and efficiency of channel transport between the proton exchange membrane and the electrode. When the temperature inside the stack is too low, the water content of the cathode will gradually accumulate until the gas transport layer and the gas flow channels are blocked. In addition, too high a water content in the fuel cell stack will also reduce the active area of the catalytic layer, which may cause the catalytic layer to be immersed in continuous dynamic operation conditions, which will greatly increase the loss of the active polarization overvoltage and the concentration polarization overvoltage, resulting in a decrease in the output voltage of the PEMFC system.
In the step 4), the screened fault diagnosis characteristic values are shown in table 2:
TABLE 2 diagnostic feature quantity
Figure BDA0004115691840000111
In the step 5), the specific contents are as follows:
the convolutional neural network is a feedforward neural network containing convolutional calculation and has a deep structure, and the traditional CNN uses multi-layer convolution and pooling to extract input object characteristics and mainly comprises an input layer, a convolutional layer, a pooling layer, a full-link layer and an output layer. The GoogLeNet convolutional neural network is a brand-new deep learning structure proposed by Christian in 2014, an initial structure is creatively designed, the depth and the width of the network are increased while the number of CNN parameters is reduced, and more fault characteristics of the proton exchange membrane fuel cell integrated system can be extracted under the same calculation amount.
The total layer number for constructing the network in the GoogLeNet is about 100 layers, the network depth is 22 layers, the main structure of the GoogLeNet is composed of 9 association modules (association modules), and the model structure is shown in figure 2. The index module assembles 3 convolution kernels of different sizes (1×1 convolution, 3×3 convolution, 5×5 convolution) and one pooling operation (3×3 max pooling) into one network module, and the whole network structure is disassembled in units of modules when designing the neural network. The input module processes the input images in four different modes, and after splicing and fusing in the feature dimension, the input images are transmitted to the next-layer association module. The convolution kernels of different sizes mean receptive fields of different sizes, and feature fusion of different scales enables feature capture of images at any position to be more efficient. When the input fault feature number is more, the direct convolution operation generates huge calculation amount, and the input feature dimension can be reduced by introducing the 1×1 convolution layer before the 3×3 and 5×5 convolution layers, so that the calculation complexity of the convolution operation is obviously reduced.
Each faulty sample is first processed by a convolution operation in the hidden layer, the output of which is defined as follows:
Figure BDA0004115691840000121
wherein f represents a nonlinear activation function; * Is a convolution operator; k is the number of convolution kernels; omega j And b j Respectively representing the weights and the bias values.
The pooling operation merges the N x 1 failed image block outputs of the previous convolutional layer into a single value of the next layer, thereby reducing the size of the feature vector. Each pooling layer typically follows the previous convolution layer, learning the edges and texture of the input fault state image using a max pooling operation, the output of which is calculated as follows:
Figure BDA0004115691840000122
wherein u (n, 1) is a window function of the previous layer of fault image block; a, a j Is the maximum value in the image block.
These local features extracted by the rolling and pooling operations are recombined by a fully connected layer connected to an output layer consisting of two neurons representing landslides and non-landslides. All parameters of the CNN layer are optimized using a back-propagation algorithm, the purpose of which is to minimize the loss value calculated from the loss function, defined as follows:
Figure BDA0004115691840000123
wherein m is the number of input landslide data, y i And
Figure BDA0004115691840000124
the true label and the predicted label of the ith input fault sample are represented respectively.
The GoogLeNet convolutional neural network is used as a pre-training model for transfer learning, the model is trained from a subset of an ImageNet large database, and the flow of the transfer learning is shown in figure 3. Pre-training model 1000 pictures can already be classified by pre-training 100 ten thousand pictures. Although no ultrasound image is contained in ImageNet, raw features such as blobs, colors, and lines extracted by the bottom convolution are present in almost all images. With the aid of transfer learning, the pre-trained network can be used to identify these original features and replace the last layers for combining the features and completing the final classification. Although the difference exists between the pre-training image and the target classification image, the convolutional neural network comprehensively trained on the large-scale high-quality ImageNet can still migrate, so that the PEMFC fault image recognition and diagnosis task is more effective.
In order to reduce redundant parameters in the original model, improve the generalization capability of the fault diagnosis model and save training time, the target network model needs to be finely tuned. Parameters of the first three subjects modules of the pre-training model are transferred to the target model and frozen. And then, the initial weights of the parameters of the six subjects modules after the pre-training model are used as initial training values of the target model, and the parameters are updated continuously in the training process. Since the pre-training network is used for classifying 1000 images, the size of the full link layer is 1×1×1000, and in order for the network to meet the requirements of the PEMFC fault feature image identification, the output size of the full link layer is changed to 1×1×5.
Example 2
The embodiment provides a fault diagnosis system of a proton exchange membrane fuel cell integrated system, which consists of a mathematical model building unit, a mathematical model verification unit, an output characteristic data acquisition unit, a characteristic image output unit and a characteristic image diagnosis unit.
Mathematical model building unit: according to an electrochemical reaction mechanism of the proton exchange membrane fuel cell and an empirical model formula under a dynamic working condition, constructing a mathematical model of a PEMFC integrated system, wherein an auxiliary machine system comprises a cooling system, an air supply system and a hydrogen supply system;
mathematical model verification unit: verifying a built mathematical model of the PEMFC integrated system according to experimental data, and analyzing the performance and power consumption of the PEMFC integrated system under a dynamic working condition;
an output characteristic data acquisition unit: obtaining output characteristic data of five states of normal, cooling system faults, hydrogen supply system faults, air supply system faults and flooding faults by changing parameters of each component of the PEMFC integrated system;
a feature image output unit: selecting some output characteristic data of the PEMFC integrated system as fault diagnosis characteristic quantity and outputting characteristic images of the selected fault diagnosis characteristic quantity;
a feature image diagnosis unit: performing weight migration on the pre-trained Googlene model by adopting migration learning to obtain an optimized Googlene convolutional neural network classification model; and finally, diagnosing the obtained characteristic images in five states by utilizing the classification model, and analyzing the result.
In the mathematical model building unit, the PEMFC integrated system mathematical model includes:
1) Fuel cell output characteristic model
The cell output voltage of the stack is expressed by the following formula:
V cell ＝E Nernst -V act -V ohm -V conc (2)
wherein V is cell Is the output voltage of the single cell; e (E) Nernst Is a thermodynamic electromotive force; v (V) act For activating polarization overvoltage; v (V) ohm Is an ohmic polarization overvoltage; v (V) conc Is concentration polarization overvoltage;
pile output voltage V stack Expressed as:
V stack ＝n·V cell (7)
wherein n is the number of cells constituting the stack;
electric energy P generated by PEMFC integrated system el_st And electric power P th_st The method comprises the following steps of:
P el_st ＝n·V cell ·i·A (8)
P th_st ＝n·(V L -V stack )·i (9)
wherein V is L I is the operating current of the electric pile, A is the surface area of the proton exchange membrane;
the electrical efficiency of the stack is calculated from the following formula:
Figure BDA0004115691840000141
wherein mu is f For the fuel utilization coefficient of the fuel to be used,
Figure BDA0004115691840000142
is the chemical coefficient of hydrogen;
2) Cooling system
The energy consumption of the cooling system is calculated as follows:
Figure BDA0004115691840000143
in xi 5 ，ξ 6 ，ξ 7 And xi 8 Is an empirical constant; f (f) w For cooling water flow, the calculation formula is as follows:
Figure BDA0004115691840000144
in xi 9 Is an empirical constant; ρ w Is the density of water;
Figure BDA0004115691840000145
is the specific heat capacity of water under normal conditions; delta T w For the temperature difference between the cooling water entering and leaving the PEMFC; p (P) humid For gas humidification energy consumption, it approximates the energy consumption of inlet temperature water into saturated steam:
Figure BDA0004115691840000146
in the method, in the process of the invention,
Figure BDA0004115691840000147
and->
Figure BDA0004115691840000148
The flow rates of the water vapor contained in the humidified air and hydrogen are respectively; />
Figure BDA0004115691840000149
Enthalpy consumed for heating 1 mole of water to water vapor at an inlet temperature of 298.15K:
Figure BDA0004115691840000151
in the method, in the process of the invention,
Figure BDA0004115691840000152
is the water molar mass under normal conditions; t (T) stack Is the temperature of the galvanic pile;
3) Air supply system
The energy consumption of the air supply system in the air intake pressurization is calculated as follows:
P comp ＝c p ·ΔT gas ·Q air (15)
wherein, c p Constant pressure specific heat capacity for inlet gas; q (Q) air Air intake mass flow; delta T gas For the temperature rise of the compressor inlet gas, the calculation method is as follows:
Figure BDA0004115691840000153
wherein p is 1 Is the pressure before the gas is compressed; p is p 2 Is the pressure of the compressed gas; gamma is the ratio of the specific heat capacity of the gas to the specific heat capacity of the constant pressure; t (T) 2 T is the temperature of the gas before being introduced into the compressor 1 The temperature of the gas after being introduced into the compressor;
4) Hydrogen supply system
The hydrogen circulation pump recovers unreacted hydrogen fuel, and the calculation result of the ideal gas adiabatic compression process is regarded as the power consumption of the circulation pump:
Figure BDA0004115691840000154
in which W is cyc Is the mass flow of the circulation pump;
Figure BDA0004115691840000155
is the specific heat capacity of hydrogen at constant pressure; t (T) t Is the gas precompression temperature; pi cyc Is an adiabatic compression ratio; η (eta) cyc Efficiency as a hydrogen circulation pump;
system electric power P ele_sys Electric efficiency eta ele_sys And system thermal efficiency eta th_sys The calculation method of (2) is as follows:
Figure BDA0004115691840000156
in the mathematical model verification unit, the performance and the power consumption of the proton exchange membrane fuel cell integrated system comprise the operating temperature, the polarization curve, the thermal efficiency and the electric power of a pile of the proton exchange membrane fuel cell integrated system.
In the output characteristic data acquisition unit, the failure generation mechanism is as shown in table 1:
TABLE 1 failure generation mechanism
Figure BDA0004115691840000157
Figure BDA0004115691840000161
In the feature image output unit, the selected fault diagnosis feature amounts are shown in table 2 below:
TABLE 2 fault diagnosis feature
Figure BDA0004115691840000162
In the characteristic image diagnosis unit, the main structure of the GoogLeNet consists of 9 association modules; the association module assembles 3 convolution kernels with different sizes and a pooling operation into a network module, and the whole network structure is assembled by taking the network module as a unit when designing the neural network; the association module processes the input images in four different modes, and after splicing and fusing in characteristic dimension, the input images are transmitted to the next-layer association module;
each faulty sample is first processed by a convolution operation in the hidden layer, the output of which is defined as follows:
Figure BDA0004115691840000163
wherein f represents a nonlinear activation function; * Is a convolution operator; n is the number of convolution kernels; omega j And b j Respectively representing the weight and the bias value; v i Is the ith characteristic image data;
pooling operation merges the n×1 faulty image block outputs of the previous convolutional layer into a single value of the next layer, thereby reducing the size of the feature vector; each pooling layer follows the previous convolutional layer, learns the edges and texture of the input fault state image using a max pooling operation, the output of which is calculated as follows:
Figure BDA0004115691840000164
wherein u (n, 1) is a window function of the previous layer of fault image block; a, a j Is the maximum value in the image block;
the local features extracted through rolling and pooling operations are recombined through a full-connection layer, and the full-connection layer is connected with an output layer, wherein the output layer consists of two neurons representing landslide and non-landslide; all parameters of the CNN layer are optimized using a back-propagation algorithm, the purpose of which is to minimize the loss value calculated from the loss function, defined as follows:
Figure BDA0004115691840000171
wherein m is the number of input landslide data, y i And
Figure BDA0004115691840000172
the true label and the predicted label of the ith input fault sample are represented respectively.
The google net convolutional neural network is used as a pre-training model for transfer learning, which is trained from a subset of the ImageNet large database, and although the ImageNet does not contain ultrasound images, the original features of the blobs, colors and lines extracted by the bottom convolution are present in almost all images, which are identified by the transfer learning by means of the pre-training network, and the last layers for combining the features and completing the final classification are replaced.
In order to reduce redundant parameters in the original model, improve the generalization capability of the fault diagnosis model and save training time, the target network model needs to be finely tuned. The fine tuning step is as follows: firstly, transferring parameters of the first three associated modules of the pre-training model to a target model and freezing; and then, the initial weights of the parameters of the six associated modules after the pre-training model are reused as initial training values of the target model, and the parameters are updated continuously in the training process.
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.

Claims (10)

1. A fault diagnosis method for a proton exchange membrane fuel cell integrated system is characterized by comprising the following steps:
1) According to an electrochemical reaction mechanism of the proton exchange membrane fuel cell and an empirical model formula under a dynamic working condition, constructing a mathematical model of a PEMFC integrated system, wherein an auxiliary machine system comprises a cooling system, an air supply system and a hydrogen supply system;
2) Verifying a built mathematical model of the PEMFC integrated system according to experimental data, and analyzing the performance and power consumption of the PEMFC integrated system under a dynamic working condition;
3) Obtaining output characteristic data of five states of normal, cooling system faults, hydrogen supply system faults, air supply system faults and flooding faults by changing parameters of each component of the PEMFC integrated system;
4) Selecting some output characteristic data of the PEMFC integrated system as fault diagnosis characteristic quantity and outputting characteristic images of the selected fault diagnosis characteristic quantity;
5) Performing weight migration on the pre-trained Googlene model by adopting migration learning to obtain an optimized Googlene convolutional neural network classification model; and finally, diagnosing the obtained characteristic images in five states by utilizing the classification model, and analyzing the result.
2. The method for diagnosing faults in a proton exchange membrane fuel cell integrated system according to claim 1, wherein in the step 1), the PEMFC integrated system mathematical model includes:
1) Fuel cell output characteristic model
The cell output voltage of the stack is expressed by the following formula:
V cell ＝E Nernst -V act -V ohm -V conc (2)
wherein V is cell Is the output voltage of the single cell; e (E) Nernst Is a thermodynamic electromotive force; v (V) act For activating polarization overvoltage; v (V) ohm Is an ohmic polarization overvoltage; v (V) conc Is concentration polarization overvoltage;
pile output voltage V stack Expressed as:
V stack ＝n·V cell (7)
wherein n is the number of cells constituting the stack;
electric energy P generated by PEMFC integrated system el_st And electric power P th_st The method comprises the following steps of:
P el_st ＝n·V cell ·i·A (8)
P th_st ＝n·(V L -V stack )·i (9)
wherein V is L I is the operating current of the electric pile, A is the surface area of the proton exchange membrane;
the electrical efficiency of the stack is calculated from the following formula:
Figure FDA0004115691830000021
wherein mu is f For the fuel utilization coefficient of the fuel to be used,
Figure FDA0004115691830000022
is the chemical coefficient of hydrogen;
2) Cooling system
The energy consumption of the cooling system is calculated as follows:
Figure FDA0004115691830000023
in xi 5 ，ξ 6 ，ξ 7 And xi 8 Is an empirical constant; f (f) w For cooling water flow, the calculation formula is as follows:
Figure FDA0004115691830000024
/>
in xi 9 Is an empirical constant; ρ w Is the density of water;
Figure FDA0004115691830000025
is the specific heat capacity of water under normal conditions; delta T w For the temperature difference between the cooling water entering and leaving the PEMFC; p (P) humid For gas humidification energy consumption, it approximates the energy consumption of inlet temperature water into saturated steam:
Figure FDA0004115691830000026
in the method, in the process of the invention,
Figure FDA0004115691830000027
and->
Figure FDA0004115691830000028
The flow rates of the water vapor contained in the humidified air and hydrogen are respectively; />
Figure FDA0004115691830000029
For inlet temperature 298.1Enthalpy consumed by heating 1 mole of water to water vapor at 5K:
Figure FDA00041156918300000210
in the method, in the process of the invention,
Figure FDA00041156918300000211
is the water molar mass under normal conditions; t (T) stack Is the temperature of the galvanic pile;
3) Air supply system
The energy consumption of the air supply system in the air intake pressurization is calculated as follows:
P comp ＝c p ·ΔT gas ·Q air (15)
wherein, c p Constant pressure specific heat capacity for inlet gas; q (Q) air Air intake mass flow; delta T gas For the temperature rise of the compressor inlet gas, the calculation method is as follows:
Figure FDA00041156918300000212
wherein p is 1 Is the pressure before the gas is compressed; p is p 2 Is the pressure of the compressed gas; gamma is the ratio of the specific heat capacity of the gas to the specific heat capacity of the constant pressure; t (T) 2 T is the temperature of the gas before being introduced into the compressor 1 The temperature of the gas after being introduced into the compressor;
4) Hydrogen supply system
The hydrogen circulation pump recovers unreacted hydrogen fuel, and the calculation result of the ideal gas adiabatic compression process is regarded as the power consumption of the circulation pump:
Figure FDA0004115691830000031
in which W is cyc Is the mass flow of the circulation pump;
Figure FDA0004115691830000032
is the specific heat capacity of hydrogen at constant pressure; t (T) t Is the gas precompression temperature; pi cyc Is an adiabatic compression ratio; η (eta) cyc Efficiency as a hydrogen circulation pump;
system electric power P ele_sys Electric efficiency eta ele_sys And system thermal efficiency eta th_sys The calculation method of (2) is as follows:
Figure FDA0004115691830000033
3. the method for diagnosing faults in a pem fuel cell integrated system of claim 1 wherein in said step 2) the performance and power consumption of the pem fuel cell integrated system include stack operating temperature, polarization curve, thermal efficiency and electrical power of the pem fuel cell integrated system.
4. The method for diagnosing a fault in a proton exchange membrane fuel cell integrated system as claimed in claim 1, wherein in the step 3),
TABLE 1 failure generation mechanism
Fault type Mechanism of production Cooling system failure (F1) Shortening coolant channel, changing laminar flow and turbulent flow Reynolds number of radiator Hydrogen supply system failure (F2) Reducing hydrogen excess ratio and intake pressure Air supply system failure (F3) Reducing air excess ratio and intake pressure Flooding failure (F4) Reducing intake relative humidity
The failure generation mechanism is shown in table 1 above.
5. The method for diagnosing a failure in a proton exchange membrane fuel cell integrated system as claimed in claim 1, wherein in said step 4),
TABLE 2 fault diagnosis feature
Figure FDA0004115691830000034
Figure FDA0004115691830000041
The selected fault diagnosis feature amounts are shown in table 2 above.
6. The method for diagnosing faults in a proton exchange membrane fuel cell integrated system according to claim 1, wherein in the step 5), the main structure of google net is composed of 9 associated modules; the association module assembles 3 convolution kernels with different sizes and a pooling operation into a network module, and the whole network structure is assembled by taking the network module as a unit when designing the neural network; the association module processes the input images in four different modes, and after splicing and fusing in characteristic dimension, the input images are transmitted to the next-layer association module;
each faulty sample is first processed by a convolution operation in the hidden layer, the output of which is defined as follows:
Figure FDA0004115691830000042
wherein f represents a nonlinear activation function; * Is a convolution operator; n is the number of convolution kernels; omega j And b j Respectively representing the weight and the bias value; v i Is the ith feature image;
pooling operation merges the n×1 faulty image block outputs of the previous convolutional layer into a single value of the next layer, thereby reducing the size of the feature vector; each pooling layer follows the previous convolutional layer, learns the edges and texture of the input fault state image using a max pooling operation, the output of which is calculated as follows:
Figure FDA0004115691830000043
wherein u (n, 1) is a window function of the previous layer of fault image block; a, a j Is the maximum value in the image block;
the local features extracted through rolling and pooling operations are recombined through a full-connection layer, and the full-connection layer is connected with an output layer, wherein the output layer consists of two neurons representing landslide and non-landslide; all parameters of the CNN layer are optimized using a back-propagation algorithm, the purpose of which is to minimize the loss value calculated from the loss function, defined as follows:
Figure FDA0004115691830000044
wherein m is the number of input landslide data, y i And
Figure FDA0004115691830000045
the true label and the predicted label of the ith input fault sample are represented respectively.
7. The method according to claim 6, wherein in the step 5), the google net convolutional neural network is used as a pre-training model for migration learning, the model is trained from a subset of the large database of imagenets, and although the imagenets do not include the ultrasound images, the original features of the spots, colors and lines extracted by the bottom convolution are present in almost all the images, and the pre-training network is used to identify the original features by migration learning, and replace the last layers for combining the features and completing the final classification.
8. The method for diagnosing a fault in a pem fuel cell integrated system according to claim 7 wherein in said step 5), fine tuning of the target network model is required to reduce redundant parameters in the original model and to increase the generalization ability of the fault diagnosis model, saving training time.
9. The method for diagnosing a fault in a pem fuel cell integrated system according to claim 8 wherein said step 5) includes the steps of: firstly, transferring parameters of the first three associated modules of the pre-training model to a target model and freezing; and then, the initial weights of the parameters of the six associated modules after the pre-training model are reused as initial training values of the target model, and the parameters are updated continuously in the training process.
10. A proton exchange membrane fuel cell integrated system fault diagnosis system, comprising:
mathematical model building unit: according to an electrochemical reaction mechanism of the proton exchange membrane fuel cell and an empirical model formula under a dynamic working condition, constructing a mathematical model of a PEMFC integrated system, wherein an auxiliary machine system comprises a cooling system, an air supply system and a hydrogen supply system;
mathematical model verification unit: verifying a built mathematical model of the PEMFC integrated system according to experimental data, and analyzing the performance and power consumption of the PEMFC integrated system under a dynamic working condition;
an output characteristic data acquisition unit: obtaining output characteristic data of five states of normal, cooling system faults, hydrogen supply system faults, air supply system faults and flooding faults by changing parameters of each component of the PEMFC integrated system;
a feature image output unit: selecting some output characteristic data of the PEMFC integrated system as fault diagnosis characteristic quantity and outputting characteristic images of the selected fault diagnosis characteristic quantity;
a feature image diagnosis unit: performing weight migration on the pre-trained Googlene model by adopting migration learning to obtain an optimized Googlene convolutional neural network classification model; and finally, diagnosing the obtained characteristic images in five states by utilizing the classification model, and analyzing the result.
CN202310218121.2A 2023-03-08 2023-03-08 Fault diagnosis method and system for proton exchange membrane fuel cell integrated system Pending CN116187195A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202310218121.2A CN116187195A (en) 2023-03-08 2023-03-08 Fault diagnosis method and system for proton exchange membrane fuel cell integrated system

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202310218121.2A CN116187195A (en) 2023-03-08 2023-03-08 Fault diagnosis method and system for proton exchange membrane fuel cell integrated system

Publications (1)

Publication Number Publication Date
CN116187195A true CN116187195A (en) 2023-05-30

Family

ID=86434467

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202310218121.2A Pending CN116187195A (en) 2023-03-08 2023-03-08 Fault diagnosis method and system for proton exchange membrane fuel cell integrated system

Country Status (1)

Country Link
CN (1) CN116187195A (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116544458A (en) * 2023-07-04 2023-08-04 国家电投集团氢能科技发展有限公司 Fault warning method and device for fuel cell system
CN117558947A (en) * 2023-11-14 2024-02-13 北京氢璞创能科技有限公司 Fuel cell on-line health diagnosis and life prediction method, device and system

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116544458A (en) * 2023-07-04 2023-08-04 国家电投集团氢能科技发展有限公司 Fault warning method and device for fuel cell system
CN117558947A (en) * 2023-11-14 2024-02-13 北京氢璞创能科技有限公司 Fuel cell on-line health diagnosis and life prediction method, device and system

Similar Documents

Publication Publication Date Title
Gu et al. Data-based flooding fault diagnosis of proton exchange membrane fuel cell systems using LSTM networks
CN116187195A (en) Fault diagnosis method and system for proton exchange membrane fuel cell integrated system
Lu et al. On-line fault diagnosis for proton exchange membrane fuel cells based on a fast electrochemical impedance spectroscopy measurement
Rotondo et al. Robust fault diagnosis of proton exchange membrane fuel cells using a Takagi-Sugeno interval observer approach
CN107180983B (en) Fault diagnosis method and system for solid oxide fuel cell stack
US11916270B2 (en) Active fault-tolerant temperature control method for proton exchange membrane fuel cell system
Steiner et al. Diagnosis of polymer electrolyte fuel cells failure modes (flooding & drying out) by neural networks modeling
Mao et al. Effectiveness of a novel sensor selection algorithm in PEM fuel cell on-line diagnosis
Liu et al. Sequence fault diagnosis for PEMFC water management subsystem using deep learning with t-SNE
Mao et al. Selection of optimal sensors for predicting performance of polymer electrolyte membrane fuel cell
Zuo et al. Data-driven flooding fault diagnosis method for proton-exchange membrane fuel cells using deep learning technologies
Zhou et al. Real-time data-driven fault diagnosis of proton exchange membrane fuel cell system based on binary encoding convolutional neural network
Wu et al. A hybrid prognostic model applied to SOFC prognostics
CN109840593B (en) Method and apparatus for diagnosing solid oxide fuel cell system failure
CN112117475A (en) Fault detection device and method for water management subsystem of fuel cell
Li et al. An LSTM and ANN fusion dynamic model of a proton exchange membrane fuel cell
CN116384823A (en) Reliability evaluation method and system for hydrogen electric coupling system
Yang et al. Fault diagnosis of SOFC system based on single cell voltage analysis
Xiao et al. Water management fault diagnosis for proton-exchange membrane fuel cells based on deep learning methods
CN114048772A (en) Fault diagnosis method, system and storage medium for fuel cell device
Lu et al. Simultaneous fault diagnosis of proton exchange membrane fuel cell systems based on an Incremental Multi-label Classification Network
Sun et al. Fault diagnosis method for proton exchange membrane fuel cell system based on digital twin and unsupervised domain adaptive learning
CN116760015A (en) Large power grid transient power angle stability evaluation method based on high-dimensional graph annotation force network
Hatti et al. Static modelling by neural networks of a PEM fuel cell
Kahia et al. Neural network based diagnostic of PEM fuel cell

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