The field of the invention

Most of the energy of the world is produced by means of oil, coal, natural gas or nuclear power. All these production methods have their specific problems as far as, for example, availability and friendliness to environment are concerned. As far as the environment is concerned, especially oil and coal cause pollution when they are combusted. The problem with nuclear power is, at least, storage of used fuel.

Especially because of the environmental problems, new energy sources, more environmentally friendly and, for example, having a better efficiency than the above-mentioned energy sources, have been developed.

Fuel cells, by means of which energy of fuel, for example biogas, is directly converted to electricity via a chemical reaction in an environmentally friendly process, are promising future energy conversion devices. The state of the art

Fuel cell, as presented in fig 1, comprises an anode side 100 and a cathode side 102 and an electrolyte material 104 between them. In solid oxide fuel cells (SOFCs) oxygen 106 is fed to the cathode side 102 and it is reduced to a negative oxygen ion by receiving electrons from the cathode. The negative oxygen ion goes through the electrolyte material 104 to the anode side 100 where it reacts with fuel 108 producing water and also typically carbon dioxide (C02). Between anode 100 and cathode 102 is an external electric circuit 111 comprising a load 110 for the fuel cell.

A solid oxide fuel cell (SOFC) device is an electrochemical conversion device that produces electricity directly from oxidizing fuel. Advantages of SOFC device include high efficiencies, long term stability, low emissions, and cost. The main disadvantage is the high operating temperature which results in long start up times and both mechanical and chemical compatibility issues. In figure 2 is presented a SOFC device as an example of a high temperature fuel cell device. SOFC device can utilize as fuel for example natural gas, bio gas, methanol or other compounds containing hydrocarbons. SOFC device in figure 2 comprises more than one, typically plural of fuel cells in stack formation 103 (SOFC stack). Each fuel cell comprises anode 100 and cathode 102 structure as presented in figure 1. Part of the used fuel can be recirculated in feedback arrangement 109 through each anode. SOFC device in fig 2 also comprises a fuel heat exchanger 105 and a reformer 107.

Typically several heat exchangers are used for controlling thermal conditions at different locations in a fuel cell process. Reformer 107 is a device that converts the fuel such as for example natural gas to a composition suitable for fuel cells, for example to a composition containing hydrogen and methane, carbon dioxide, carbon monoxide and inert gases. Anyway in each SOFC device it is though not necessary to have a reformer. By using measurement means 115 (such as fuel flow meter, current meter and temperature meter) necessary measurements are carried out for the operation of the SOFC device. Part of the gas used at anodes 100 may be recirculated through anodes in feedback arrangement 109 and the other part of the gas is exhausted 114 from the anodes 100.

SOECs (Solid Oxide Electrolyzer Cell). SOEC is a solid oxide cell that comprises similar structure and operation principles as SOFC, but SOEC is operated to opposite direction as SOFC to achieve the electrolysis of water and/or carbon dioxide as fuel. SOEC uses a solid oxide, or ceramic, electrolyte to produce oxygen and hydrogen gas. In regenerative mode same cell or same cell stack can be operated to both operation directions, i.e. to the SOFC operation direction or to the SOEC operation direction. Several stacks are needed in new technology solid oxide systems 20kW or more, and thus comparisons and diagnostics between the stacks are more and more important. Power plant processes, such as fuel cell power plant, can typically be modelled with good accuracy, and such modelling

simulations can be used e.g. for sizing purposes, process analysis and optimization, and also for making on-line diagnostics of cell system

conditions. Simulations can be based e.g. on solving physical equations, semi-empirical correlations, neural network modelling, and combinations of these. Comparison of simulated model output and actual measured output of solid oxide fuel cell stack (SOFC) in SOFC based power plant could be used to indicate faults in the SOFC system and in the stack. Possible output parameters for SOFC measurement and model are anode and cathode outlet gas temperatures (T_air_out, T_fuel_out) and SOFC voltage (V). Sudden increased difference between model and measurement in these parameters could indicate e.g. mechanical deformation of the SOFC stack or problem in balance-of-plant (BoP) component affecting stack input parameters.

Two problems relate to model based SOFC diagnostics 1) During the lifetime of SOFC power plant the SOFC stacks degrade and their input-output relation changes gradually, which is very difficult to model accurately due to complex nature of cell and stack degdaration phenomena. 2) Each SOFC stack is an individual making the use of general stack model impossible without jeopardizing the sensitivity of model based diagnostics. Using a neural network based stack model and following procedures these problems can be overcome. These problems can also be overcome with physical equation based models with proper parameters that can be adjusted during operation.

Using a neural network model consists of the training of the model and the actual usage of the model. During training of the model set of measured input and output parameters are given to the model and the model is fitted to correspond them. During usage of the model measured input parameters are given to the model and simulated model output is compared to the measured output.

In patent document EP1408384B1 (ST MICROELECTRONICS SRL [IT]) is presented an arrangement for controlling a system, such as a fuel cell system, according to the deviation (ERR) between the value measured on the system (VR) and the value (VS) estimated by means of a model of the controlled system (S) of at least one control parameter. Neural network generates the estimation (VS) of said control parameter implementing said model as a function of a set of characteristic parameters of the controlled system (S) and of respective configuration parameters of the neural network. Said neural network is associated thereto a training module, which can train the neural network by modifying the configuration parameters according to a set of updating data. The embodiments presented in patent document EP1408384B1 can not model or control separately operation of separate fuel cell stacks in the system or separately operation of separate groups of stacks in the system. Instead of that in EP1408384B1 is focused on controlling the whole system as a control entity, and not making difference between separate fuel cell stacks. A defect of embodiments in EP1408384B1 is also that fault diagnosis and updating of stack models can not be fluently performed during operation of the fuel cell system.

Short description of the invention The object of the present invention is to accomplish an indication

arrangement for a solid oxide cell system, in which arrangement operation conditions of cell stacks or groups of cell stacks can be individually followed, and the indication arrangement can also be individually and automatically updated during the operation of the solid oxide cell system. This is achieved by an arrangement for diagnosis of solid oxide cells operating conditions in a solid oxide cell system, each cell in the cell system comprising an anode side, a cathode side, and an electrolyte between the anode side and the cathode side, the cell system comprising the cells in cell stacks, air feed-in piping for feeding air to the cell stacks, and fuel feed piping for feeding fuel to the cell stacks. The arrangement comprises means for determining at least one of stack specific and group of stacks specific operation window of the solid oxide cell system stacks on the basis of stack information or group of stacks information comparisons, means for performing real time stack modelling of the solid oxide cell system individual stacks or groups of stacks essentially simultaneously during operation of the solid oxide cell system to define as monitored parameters at least one of stack voltage value, air output temperature value, internal temperature value of stack, fuel output temperature value and leakage rate on the basis of input parameters of the solid oxide cell system operation provided to the stack modelling, means for monitoring, if at least one deviation of said monitored parameters exceeds the determined operation window, and means for performing fault

diagnostics, when the operation window has been monitored to be exceeded.

The focus of the invention is also a method for diagnosis of solid oxide cells operating conditions in a solid oxide cell system, wherein cells being formed in cell stacks, air being fed into the cell stacks and fuel being fed to the cell stacks. In the method is determined at least one of stack specific and group of stacks specific operation window of the solid oxide cell system stacks on the basis of stack information or group of stacks information comparisons, is performed real time stack modelling of the solid oxide cell system individual stacks or groups of stacks essentially simultaneously during operation of the solid oxide cell system to define as monitored parameters at least one of stack voltage value, air output temperature value, internal temperature value of stack, fuel output temperature value and leakage rate on the basis of input parameters of the solid oxide cell system operation provided to the stack modelling, and in the method is monitored, if at least one deviation of said monitored parameters exceeds the determined operation window, and is performed fault diagnostics, when the operation window has been monitored to be exceeded. The invention is based on determination of at least one of stack specific and group of stacks specific operation window of the solid oxide cell system stacks on the basis of stack information or group of stacks information comparisons, and on real time stack modelling of the solid oxide cell system individual stacks or groups of stacks essentially simultaneously during operation of the solid oxide cell system on the basis of input parameters of the solid oxide cell system operation, said input parameters being provided to the stack modelling. The comparisons performed between stack information or group of stacks information make possible to detect a deviation in operation of an individual stack or group of stacks at a very early stage.

The benefit of the invention is that operation deviations of individual stacks or group of stacks can be detected earlier than with traditional

measurements, and necessary repairing action and/or preventive

maintenance can be performed immediately to prevent defects from developing further in the stacks. Thus the benefits of the invention also includes cost savings and an increase in reliability and life time of the solid oxide cell system.

Short description of figures

Figure 1 presents a single fuel cell structure. Figure 2 presents an example of a SOFC device.

Figure 3 presents a preferred embodiment according to the present

invention. Detailed description of the invention

Solid oxide fuel cells (SOFCs) can have multiple geometries. The planar geometry (Fig 1) is the typical sandwich type geometry employed by most types of fuel cells, where the electrolyte 104 is sandwiched in between the electrodes, anode 100 and cathode 102. SOFCs can also be made in tubular geometries where for example either air or fuel is passed through the inside of the tube and the other gas is passed along the outside of the tube. This can be also arranged so that the gas used as fuel is passed through the inside of the tube and air is passed along the outside of the tube. Other geometries of SOFCs include modified planar cells (MPC or MPSOFC), where a wave-like structure replaces the traditional flat configuration of the planar cell. Such designs are promising, because they share the advantages of both planar cells (low resistance) and tubular cells.

The ceramics used in SOFCs do not become ionically active until they reach a very high temperature and as a consequence of this the stacks have to be heated at temperatures ranging typically from 600 to 1,000 °C. Reduction of oxygen 106 (Fig. 1) into oxygen ions occurs at the cathode 102. These ions can then be transferred through the solid oxide electrolyte 104 to the anode 100 where they can electrochemically oxidize the gas used as fuel 108. In this reaction, water and carbon dioxide byproducts are given off as well as two electrons. These electrons then flow through an external circuit 111 where they can be utilized. The cycle then repeats as those electrons enter the cathode material 102 again.

The present invention can be utilized in solid oxide cell systems such as SOFCs or in SOECs (Solid Oxide Electrolyzer Cell) presented in the state of the art. The invention can also be utilized in other systems such as in MCFCs (Molten Carbonate Fuel Cells) and other high temperature fuel cells that operate at 400 °C and higher temperatures. MCFCs are high-temperature fuel cells that use an electrolyte composed of a molten carbonate salt mixture suspended in a porous, chemically inert ceramic matrix of BASE, Beta-Alumina Solid Electrolyte.

In an arrangement according to the present invention can be utilized preferred input parameters from system measurements and calculations from solid oxide cell system automation to means 122 (fig. 3) for performing real time stack modelling of the solid oxide cell system individual stacks 103 or groups of stacks 103. Said preferred input parameters are the following: stack current (I) value, air utilization (AU) rate, air inlet temperature (T _aiUn) value, fuel utilization (FU) rate and environment temperature (T _env) value. Especially when the system is a fuel cell system a preferred input parameters are fuel inlet concentration values such as for example methane (CH4) content value, hydrogen (H2) content value and water (H20) content value. Said input parameters to calculation processing of the means 122 (i.e. preferably to a digital processor) are to define at least one of cell voltage value, internal temperature value of stack, fuel output temperature value, leakage rate and air output temperature value., At least the cell voltage value is preferably stack model specific or group of stacks model specific.

In order to generate a training dataset for neural network real time stack modelling by the means 122, the input parameters need to be varied. They can be varied for example by varying stack current (I, which affects directly or indirectly to all of these), reformer inlet temperature (T _ref, affects concentrations) value and air inlet temperature (T _aiUn) value, which affects also environment temperature (T _env) value. By varying I, T_ref and T_aiUn simultaneously in the solid oxide cell system, map of combinations of stack (or stack group) model input parameters will be gone through and training set for neural network stack model or for stack group model is generated. Stack comprises at least one cell, preferably several cells. Stack group comprises more than one cell stack, and the stacks in said group are for example in electrical serial connection to each other or mechanically connected to each other. The scale and sufficiency of said map can be pre- evaluated with solid oxide cell system model, and also a processor tool used for said generation of training datasets for the models can be pre-trained with said pre-evaluation with the solid oxide cell system model.

In another embodiment real time stack modelling is performed by the means 122 on the basis of at least one physical model. The physical model can be fitted on the basis of a difference value between a measured voltage value and a voltage value given by the model to fit an ASR (Area Specific

Resistance) parameter value and/or to fit a leakage parameter value. ASR and leakage value are typically the changing parameters due to degradation and use of the cell stacks 103. ASR parameter value and/or leakage parameter value can also be fitted on the basis of said difference voltage value and difference value between gas output temperatures to change ASR and/or leakage value so that said difference values would adapt to minimum values or even to zero. In the physical stack model embodiment it is possible to perform same kind of varying of input parameters and updating of models as in the neural network stack model embodiment. In figure 3 is presented a preferred embodiment according to the present invention, where is arranged an automatic and adaptive indicator of operating conditions of solid oxide fuel cells. The solid oxide cell system in this exemplary preferred embodiment is a solid oxide fuel cell system

(SOFC). The arrangement comprises means 130 for determining at least one of stack specific and group of stacks specific operation window of the solid oxide cell system stacks 103 on the basis of stack information or group of stacks information comparisons. These comparisons are performed by comparing at least one of comparisons individual stack 103 to individual stack 103 or to individual stacks 103 value information, individual stack 103 to average stack value information or to average group of stack, in which the individual stack is situated, value information, group of stacks 103 to group of stacks 103 or to groups of stacks 103 value information and group of stacks 103 to average group of stacks 103 value information in order to detect differences between stacks 103 or between groups of stacks 103. Preferably said comparisons are performed by same means 130, which also determine operation window(s). Said means 130 are for example a computer processor or some other processor, which performs said comparisons on the basis of calculative results and measurement results of the SOFC stack(s) 103.

By this way a deviation of an individual stack is detected so early that prior art measurements would not have been indicating anything yet. Said detected deviation indicates for example of an increased leakage level of said stack, of potential problem in fuel delivery distribution, or of a stack which is degrading faster than normally due to some reason. By this way is detected also a deviation of an individual mechanically and/or electrically connected group of stacks compared to at least one other group. Said detected deviation indicates for example of potential problem in fuel delivery distribution or of an increased leakage level of an individual stack, which causes increase in temperature in surroundings of the group of stacks. Thus the increased leakage level of one stack in the group may affect to other stacks in said group. An indication can also be an increase in varying of temperature distribution of a stack module caused by a normal degradation. By this way is detected e.g. a deviation between symmetrically connected groups of stacks. Said detected deviation indicates for example of potential problem in fuel delivery distribution because delivery pipings are typically never fully symmetrical. Asymmetry causes distortion in temperature delivery of the stacks during degradation process. Said deviation can also indicate for example of leakage in pipings or of coking causing additional flow resistance in the pipings where it locates. On the basis of said detected deviations and comparisons between different stacks and stack groups can thus be determined what is the problem causing the deviations and repairing actions can be taken. Said actions can be for example current modulation of the group of stacks which have detected problem in fuel delivery distribution or preventive maintenance if an indication of leakage in the piping is detected.

The embodiment according to figure 3 also comprises means 122 for performing real time stack modelling of the solid oxide cell system individual stacks 103 or groups of stacks 103 essentially simultaneously during operation of the solid oxide cell system to define at least one of stack voltage value, air output temperature value, internal temperature value of stack, fuel output temperature value and leakage rate on the basis of input parameters of the solid oxide cell system operation provided to the stack modelling. The internal temperature value of stack can be at least of internal minimum temperature value of stack, internal maximum temperature value of stack, internal maximum temperature gradient of stack and internal temperature profile of stack.

Said real time modelling is preferably neural network modelling or modelling on the basis of at least one physical model performed in a calculative process by the means 122, which are preferably a digital processor, such as a computer unit 122. Each stack is modelled and/or stacks are grouped and each group of stacks is modelled separately. Said input parameters of the solid oxide cell system operation are one or more of individual stack current value, air utilization rate, air flow rate, air inlet temperature value, fuel utilization rate, fuel flow rate, fuel composition information, cell system surroundings temperature value, heat flux to surroundings, and some other possible input parameters. The input parameter fuel composition information can be provided as fuel flow information, which is, when needed,

supplemented with pre-reformer outlet gas temperature information. The input parameter fuel composition information can be also provided as at least one or more of methane CH4, hydrogen H2, water H20, carbon monoxide CO, carbon dioxide CO2, nitrogen N2, or other species content information. By using variation of one or more of said input parameters is formed training stack model information to be utilized in updating of the stack modelling. The arrangement of figure 3 comprises means for providing input parameters by measurements from the solid oxide cell system and/or by calculations in the computer 122 processor. Said stack modelling is performed essentially simultaneously during operation of the solid oxide cell system.

The preferred embodiment according to the invention comprises means 132 for monitoring, if the determined operation window is exceeded on the basis of a detected deviation, and means 128 for performing fault diagnostics, when the operation window has been monitored to be exceeded. Both said means 132, 128 can be realized for example in a computer processor. The preferred embodiment in figure 3 shows an example that all these processor based means 122, 128, 130, 132 locate in a same computer unit. The stack models of the individual stacks 103 or groups of stacks 103 are updated to correspond to a new operation window determined on the basis of said fault diagnosis.

In one embodiment of the invention the real time stack modelling can be performed, for example in a computer processor 122, by fitting a physical model of the solid oxide cell system individual stacks 103 or groups of stacks 103. Said fitting of the physical model can be made on the basis of a difference value between a measured voltage value and a voltage value given by the model to fit an ASR (Area Specific Resistance) parameter value of the fuel cell(s) and/or leakage parameter value . Means 122, 128, 130, 132 can locate in one or more of the following locations: solid oxide cell system automation hardware, an external computer and a centralized server, in which remote simulation is performed on the basis of data logged therein. The centralized server provides a possibility to utilize data from several solid oxide cell systems.

In the following description is described more detailed information about neural network stack modelling characteristics according to the present invention. Dependent on the application need, in the neural network stack modelling can be modelled individual stack model, model of group of stacks, model of average stack in stack group, average stack in a stack module, i.e. in a power module, which comprises preferably of several stacks or stack groups etc. In the fault analysis according to the embodiments of the invention can be made comparisons between different detail levels e.g.

comparisons between individual stacks, between groups of stacks, between two stacks in one bundle, and between an average stack in unit versus average stack in all units.

Detection of immediate changes can be based on static comparisons, when deviation between stack model output voltage value and measured stack voltage output value exceeds tolerance limit output model voltage - output measured voltage > tolerance limit. In detection of time-dependent change in deviation between stack model voltage value and measured stack voltage value is followed e.g. if the change in deviation is faster than expected degradation rate or faster than expected increase in degradation rate.

Control means of the solid oxide cell system are used to take control over the detected deviation preferably on the basis of diagnosis telling nature of the actual problem indicated by said detected deviation. In the case of an individual stack or group of stacks the control action can be performed for example by current modulation and in the case of several stack groups the control action can be performed for example on the basis of modifying fuel utilization of the solid oxide cell system.

In the neural network stack modelling according to the present invention can be detected for example mechanical deformation of stack, changes in stack leakages, degradation of stack, damaged stack cell material and coking. Also in said stack modelling can be detected faults in solid oxide cell system measurements and faults in balance-of-plant components affecting non- measured stack model input parameters, e.g. reformer composition is not in equilibrium, recycling rate of anode-loop, etc. The following actions can be taken when stack modelling, for example neural network stack modelling, is detected to need updating: 1. Automatic or manual fault diagnostic protocol is run. 2. After the error situation is dealt, stack model(s) are updated to correspond the new situation preferably on the basis of the neutral network training model dataset(s) or on the basis of physical model based training model dataset(s).

In embodiments according to the invention the stack comprises of one or more cells. This means that in one embodiment a stack is a cell. Instead of modelling individual stacks or stack groups also individual cells or cell groups can be modelled to take same kind of method steps as described in the detailed description of the invention.

Although the invention has been presented in reference to the attached figures and specification, the invention is by no means limited to those as the invention is subject to variations within the scope allowed for by the claims.