TECHNICAL FIELD

[ 0001 ] The present invention refers to a method of monitoring a facility containing a continuous flow engine or a continuous device providing an improved monitoring even for di fficult systems being highly complex . Furthermore , the present invention refers to monitoring system adapted to be utili zed in such facility . Additionally, the present invention refers to a computer program product to reali ze such method .

BACKGROUND

[ 0002 ] The industrial sector utili zes a plurality of devices . Many to most of the devices are utili zed batchwise . However, in many industrial fields devices operate continuously essentially represent the backbone the corresponding industrial field depends upon . Compared to the batchwise utili zed devices such continuously utili zed devices providing an uninterrupted operation over months and longer provide di f ferent possibilities and challenges . A batchwise utili zed device can be , for example , checked and restarted again and again . However, such continuously operated devices do not allow to simply be turned of f without good reason . The continuous operation typically results in a plurality of data and experience acquired over years and decades . Allowing the skilled person working with such device to early detect minor changes to predict problems and correctly react to the speci fic needs at a certain point of time . Yet , upgrades and a more speci fic adaption according to the very speci fic needs and application result in constant changes making it hard for even experienced skilled persons to keep up to date . Leaving alone new generation of experts being confronted with an overwhelming amount of collected data and, for example , the task to correctly j udge whether some historic data is relevant or not for the speci fic application . Taking not only into account the speci fic historic data, but also the maintenances and upgrades done in the meantime .

[ 0003 ] For example , continuous flow engines are very important and highly sophisticated devices utili zed in modern industry . They are utili zed in a plurality of applications and ful fills their tasks over years and decades . Being subj ect to maintenance and upgrades , but essentially remain the same . While the control of such continuous flow engines is naturally well developed, they are constantly evolving with even increased speed today . New available possibilities like additive manufacturing enabling to provide speci fically tailored components in low numbers ef ficiently provide the possibility and challenge to even more adapt such continuous flow engines to the very speci fic application . Allowing to constantly improve the benefit obtained herewith . However, simultaneously confronting the persons working with such continuous flow engine to solve the above-described challenges . Either resulting in an increased risk of damages or dangers or a more restrictive utili zation rendering the possible benefit of , for example , upgrades to be quite limited or worst case even lowering the output of such devices below the pre-upgrade values .

[ 0004 ] The field of power generation and distribution is a known example of conservative device handling, as reliability and security are top targets to be secured . For example , the circuit breakers and electric grids utili zed in such field are essentially constantly operating . Typically, no replacement is available or some local solution to reroute some electricity or utili ze some movable solution is only available for short time and providing such means typically requires extensive time and ef fort to prepare so . Identi fying a problem in time to include it in some scheduled yearly maintenance or the like is a demanding task . And whether some deviation from the typical behavior can be ignored, might possibly be resulting from a past upgrade from some device within a managed electrical grid, or requires some immediate action to safely turn of f the device before a grave danger for people is occurring .

SUMMARY

[ 0005 ] This and further problems are solved by the products and methods as disclosed hereafter and in the claims . It was noted that the above referenced problems are especially emphasi zed in the energy production related industry and the utilization of continuous flow engines , naturally especially including the utili zation of continuous flow engines in the field of energy production . Such applications can be surprisingly well supported by the solution as speci fied hereafter . Further beneficial embodiments are disclosed in the dependent claims and the further description and figures . These benefits can be used to adapt the corresponding solution to speci fic needs or to solve additional problems .

[ 0006 ] According to an aspect the present invention refers to a method of monitoring an industrial plant containing at least one continuous flow engine providing an current state , a power plant containing at least one continuous device providing a current state or a power distribution facility containing at least one continuous device providing a current state , wherein the method utili zes a monitoring device , wherein the monitoring device contains an plurality of software agents , wherein each of the plurality of software agents contains an arti ficial intelligence , wherein each of the plurality of software agents monitors a segregated part of the industrial plant , wherein the segregated parts of the industrial plant are speci fied based on knowledge graph layers stored in a knowledge graph database , wherein the monitoring device receives sensor data of sensors contained in the industrial plant , wherein the sensor data is related to the at least one continuous flow engine or the at least one continuous device , wherein the plurality of software agents monitor the at least one continuous flow engine or the at least one continuous device based on the sensor data, wherein the software agents work together to identi fy the current state and changes of the at least one continuous flow engine or the at least one continuous device .

[ 0007 ] It should be expected that splitting such monitoring into a plurality of software agents is detrimental based on conflicting evaluations and data provided . This should render such system inef ficient or worthless for applications like continuous flow engines requiring a highly reliable and quick assessment of the current situation . Yet , the inventors found that the inventive utili zation of a plurality of software agents not only provides the required reliability and speed of processing . The split into di f ferent interacting software agents , for example , additionally provides the benefit of a simpli fied provision of the required arti ficial intelligences . For example , it was noted that the training of corresponding arti ficial intelligences and eventually adapting their interaction is signi ficantly simpler than providing a comparable ef ficient and reliable generic arti ficial intelligence for monitoring such highly complex devices like continuous flow engines . While the inventive method easily provides the required reliability and further requirements in this context , the split into di f ferent software agents provides a plurality of additional benefits . For example , the overall ef ficiency and required processing power is signi ficantly decreased . Also , the possibility to subsequently adapt the single software agents allows to very easily adapt such system flexibly to changes or very speci fic needs . Not only simpli fying the monitoring of corresponding devices . But also enabling to provide highly tailored and easily updated arti ficial intelligence-based systems .

[ 0008 ] The phrase "the sensor data is related to the at least one continuous flow engine" refers to sensor data being directly or indirectly related to the at least one continuous flow engine . For example , such sensor data being directly related to the at least one continuous flow engine is sensor data of the combustion temperature of a gas burner or rotation speed of some rotor of a gas turbine . For example , such sensor data being indirectly related to the at least one continuous flow engine is sensor data of auxiliary devices like devices providing cooling air for cooling vanes and blades of a gas turbine or a power output of electricity generated by a gas turbine .

[ 0009 ] The term "power plant" as used herein refers to classical power producing facilities like power plants utili zing steam turbines and gas turbines as well as wind powerbased power producing facilities and solar power-based facilities .

[ 0010 ] The term "continuous device" as used herein refers to a device operating continuously for a prolonged period of time generating a continuous stream of data related to its operation . Such prolonged period is preferable at least one month, more preferred at least 3 months , even more preferred at least 1 year . Examples of such continuous devices are circuit breakers as used in power distribution, hydrolyzers as utili zed to convert renewable energies into hydrogen, or turbines utili zed in wind power-based power plants .

[ 0011 ] The t erm "continuous flow engine" as used herein refers to a device utili zing a continuous stream of a fluid like a gas or a liquid . Herein, such continuous flow engine typically provides a rotor located in the fluid and interacting with said fluid . Herein, such fluid can either be utili zed to provide a rotational movement of the rotor being able to be trans formed into , for example , electrical energy . Examples of such continuous flow engines are gas turbines and steam turbines . Alternatively, the rotor can actively be rotated allowing to , for example , compress the fluid . An example of such application is a compressor as utili zed, for example , in oil refineries . [ 0012 ] The phrase "comparable continuous flow engine" refers to a continuous flow engine providing comparable technical properties compared to the continuous flow engine in question . For example , it refers to other continuous flow engines of the same model series . Such same model series of continuous flow engines are well established based on the broad application of continuous flow engines in the art . Like the SGT- 800 series of Siemens representing a model series of gas turbines being distributed in many countries and well established to provide a reliable backbone of the power generation .

[ 0013 ] The phrase "comparable continuous device" refers to a continuous device providing comparable technical properties compared to the continuous device in question . For example , it refers to other continuous devices of the same model series .

[ 0014 ] The term "current state" as used herein refers to the current state the corresponding device like the continuous flow engine is operating . The term current state in this context comprises the operating state of such device as well as the operational data and state transitions . Typically, it is preferred that the current state especially contains the state transitions . In typical application cases the current state at least contains two , more preferred all three , of the aforementioned information operational data in this context refer to conditions like oil pressure , power output , cooling air temperature , or the like .

[ 0015 ] Herein, the databases can be located on a single server . However, it is typically preferred that the databases are located at di f ferent locations . For example , it is typically preferred to locate the operator database and the simulation database at di f ferent locations . While this requires additional ef fort it allows to make best use of the typically split expertise and processing power required as the simulation database is typically preferably provided by a third party investing in extensive ef fort and processing power to provide such database for a big number of users . While the practical experience showed that the users utili zing such continuous flow engines typically are reluctant to share insight in their speci fic operation and want to keep this expertise internal . Resulting in such method of operation and providing correspondingly adapted monitoring devices being surprisingly beneficial .

[ 0016 ] The term "monitoring device" as used herein refers to a physical device containing at least one processing unit and a data storage . While it is possible to place the parts of such device at di f ferent locations inside the industrial plant , they all work together and are managed together .

[ 0017 ] According to a further aspect the present invention refers to a monitoring device adapted to reali ze an inventive method, wherein the monitoring device contains at least one data storage and at least one processing unit , wherein the at least one data storage contains the plurality of software agents .

[ 0018 ] According to a further aspect the present invention refers to an upgrade kit containing an inventive monitoring device , wherein the upgrade kit is adapted to replace a conventional monitoring device without an arti ficial intelligence . It was noted that the inventive monitoring device allows to easily replace an existing monitoring device not providing a signi ficant amount of processing power . In fact , the very ef ficient monitoring system as disclosed herein allows to provide an upgrade kit including the required processing power to upgrade old monitoring systems of existing industrial plants like power plants . The low amount of processing power allows to bundle corresponding processing units in a package to be introduced in such existing systems with the available space and even power sources typically available . On the contrast , it was noted that monitoring systems containing a single arti ficial intelligence , for example , additionally required a far bigger amount of processing power requiring in turn a signi ficantly larger amount of space typically requiring an adaption of corresponding monitoring and control arrangements as well as additional power supply to be provided to ful fill the requirements .

[ 0019 ] According to a further aspect the present invention refers to a computer program product , tangibly embodied in a machine-readable storage medium, including instructions operable to cause a computing entity to execute an inventive method .

[ 0020 ] According to a further aspect the present invention refers to a storage device for providing an inventive computer program product , wherein the device stores the computer program product and/or provides the computer program product for further use .

[ 0021 ] To simpli fy understanding of the present invention it is referred to the detailed description hereafter and the figures attached as well as their description . Herein, the figures are to be understood being not limiting the scope of the present invention but disclosing preferred embodiments explaining the invention further .

BRIEF DESCRIPTION OF THE DRAWINGS

[ 0022 ] Fig . 1 shows a generic scheme of an inventive method .

DETAILED DESCRIPTION

[ 0023 ] Preferably, the embodiments hereafter contain, unless speci fied otherwise , at least one processor and/or data storage unit to implement the inventive method .

[ 0024 ] Unless speci fied otherwise terms like „calculate" , „process", "determine" , „generate" , „configure" , "reconstruct" and comparable terms refer to actions and/or processes and/or steps modi fying data and/or creating data and/or converting data, wherein the data are presented as physical variable or are available as such .

[ 0025 ] The term „data storage" or comparable terms as used herein, for example , refer to a temporary data storage like RAM (Random Access Memory) or long-term data storage like hard drives or data storage units like CDs , DVDs , USB sticks and the like . Such data storage can additionally include or be connected to a processing unit to allow a processing of the data stored on the data storage .

[ 0026 ] According to one aspect the present invention refers to a method as described above .

[ 0027 ] According to further embodiments it is preferred that the inventive method is utili zed for continuous flow engines . It was noted that the application of the invention in such area was especially beneficial . The one the one hand rely on expertise of experts to ensure a safe and reliable handling . Furthermore , they represent a highly conservative field based on the huge demands from power security and the like . On the other hand, they are currently in the process of implementing modern technologies to meet the continuously increasing demands . While being suf fering from the related problems and challenges . Corresponding continuous flow engines are , for example , typically utili zed as base power providing units in a power generation and distribution network additionally having to handle the fluctuations resulting from the inhomogeneous power generation resulting from renewable energy .

[ 0028 ] According to further embodiments it is preferred that each of the at least one continuous flow engine or each of the at least one continuous device is monitored by at least 20 , more preferred at least 50 , even more preferred at least 100 , software agents . It should be expected that utili zing a signi ficant number of arti ficial intelligences to monitor a single continuous flow engine or continuous device should result in additional problems based on conflicting observations and comparable problems . However, it was surprisingly noted that utili zing a big number of the inventive software agents in such way is very beneficial . Splitting the monitoring of such complex device in such way not only increases the processing speed and reduces the required processing power . It also allows to tailor the monitoring device easily according to the speci fic needs and improve the possibility to understand the process of evaluating and decision and simpli fies to enhance such system in the future .

[ 0029 ] According to further embodiments it is preferred that the segregated parts of the industrial plant are at most partially overlapping . At most partially overlapping in this context has the meaning that the evaluated segregated parts are not identical or one is contained within the other .

Herein, it is explicitly not excluded that in addition to the speci fied software agents such doubled software agents or overarching software agents are included . However, the specified software agents refer to the mesh of software agents providing together the monitoring of the respective device . Typically, it is preferred that at least 80% , more preferred at least 90% , even more preferred all , of the segregated parts of the industrial plant do not overlap . It is possible that some segregated parts of the industrial plant overlap to a certain degree or in certain cases completely for speci fic applications . Enabling the operator to monitor certain parts of the industrial plant by several software agents . Additionally, sensor data of a speci fic element of the industrial plant can be relevant for di f ferent evaluations . For example , for gas turbines a sensor relating to the fuel flow might be relevant to the fuel amount monitoring as well as , for example , a monitoring of the burner characteristics indicating a misbehavior of the burner in case the flame characteristics like temperature does not correlate to the fuel flow . Indicating, for example , that some clogging or damage of the burner at the mixing area of fuel and air or the like has taken place . Being veri fied by corresponding data of other software agents to real time detect corresponding development and act before problems occur .

[0030 ] According to further embodiments it is preferred that additionally to the plurality of software agents monitoring segregated parts of the industrial plant the method of monitoring utili zes doubled software agents and/or overarching software agents , wherein each of the doubled software agents and/or overarching software agents contains an arti ficial intelligence . The term "doubled software agent" as used herein refers to a software agent monitoring the same segregated part of the of the industrial plant . The term "overarching software agent" refers to a software agent monitoring a part of the industrial plant containing the segregated parts of multiple of the plurality of software agents . For many applications it is preferred that such overarching software agents refers to segregated parts of the industrial plant relating to the same continuous flow engine or continuous device . However, it was noted that is can also be preferred to include overarching software agents monitoring segregated parts of di f ferent continuous flow engines or continuous devices . Especially, it was noted that this allows to compare comparable segregated parts of di f ferent continuous flow engines or continuous devices to very easily identi fy problems no relating to a single device . Structuring the monitoring accordingly was not to allow even detecting problems not relating to the single device . Like an incorrect quality of the fuel resulting in a misbehavior of multiple gas turbines being utili zed parallel . Including doubled software agents was noted to enable to further tailor the monitoring very easily to the speci fic needs . For example , it allows to assign the feedback of an operator to the output of such doubled software agents being trained di f ferently to automatically set priority to a speci fically trained software agent . Or to provide di f ferent levels of output like a first indication of a problem and a second high alert output to immediately take action . Of fering a wide range of increased possibilities on how to improve the monitoring the operator can decide on and easily implement them . [ 0031 ] According to further embodiments it is preferred that the plurality of software agents operates based on the combination of arti ficial intelligences and predefined rules . Typically, it is preferred that at least 50% , more preferred at least 75% , even more preferred all , of the software agents operate based on a combination of an arti ficial intelligence and predefined rules . Such predefined rules can be amended by replacing existing rules , can be rolled out as batch update to update , for example , at least 10% of the software agents simultaneously, and/or can be retrieved as copy from the monitoring device , amended and uploaded again into the monitoring device . Such rules can, for example , speci fy requirements on additional software agents veri fying an alert or a more preemptive alert even in case certain signals are not received . Also , for example , that certain software agents are not allowed to trigger certain alerts . Allowing to , for example , manually adapt the operation of the plurality of software agents . Herein, it was noted that such rules can typically be simply maintained after retraining of the software agents or the like . While the software agents can be changed their interaction and correlation typically remains the same . And, for example , a rule that certain alerts of speci fic software agents need to be double checked or require the correlation with a further software agent . Herein, the additional ef fort for generating such rules is easily compensated by the benefit obtained herewith and the possibility to reuse them after replacing or retraining the software agents . While it was noted that such possibility is highly desirable in real applications . Like some repeated incorrect alert as some persons working on the device are routinely shutting of f certain parts during routine actions resulting in incorrect sensor data and correspondingly misinterpretations . While such thing should be prevented in general a more realistic approach for unproblematic cases can be to define corresponding rules to prevent incorrect alerts . Naturally, the possibility further tailor the alerts accordingly are also highly desirable . [ 0032 ] According to further embodiments it is preferred that the software agents identi fy failure modes and correspondingly label events accordingly and/or provide solutions for such failures . It was noted that training corresponding arti ficial intelligences can be reali zed while , for example , include simulated failure modes . Correspondingly trained artificial intelligences can reliably identi fy corresponding failures based on the interaction of the plurality of software agents working together and simultaneously identi fy a failure from multiple perspectives . While a single software agent might incorrectly report such failure or might not identi fy the speci fic failure it was noted that the inventive method allows to reali ze it even for complex systems like continuous flow engines . It was especially noted that splitting up the task between di f ferent software agents each being able to be monitored with regard to its reliability of assessment and quality of evaluations and suggestions created allows to very easily, reliably, and swi ftly determine the quality of corresponding data provided by the monitoring device . Allowing to utili ze the inventive method for even sensitive parts of the operation of the at least one continuous flow engine or the at least one continuous device easily leading to grave problems or even signi ficant dangers . For example , it was noted that many failure modes provide a very characteristic fingerprint at a very early stage , wherein certain software agents reporting certain problems in a certain order . Allowing to signi ficantly improve the prediction of problems at an early stage .

[ 0033 ] According to further embodiments it is preferred that the method contains providing evaluation data regarding an identi fication of a failure or a solution provided for such failure , wherein the evaluation data contains data how the software agents identi fied the failure and/or provided the solution . For example , it was surprisingly beneficial and easy to be reali zed to note down the software agents involved in the identi fication and/or solution generation . While it is typically preferred to include additional data, it was noted to already provide a signi ficant benefit and insight in corresponding evaluation processes simply to know such involved software agents . Allowing to easily provide a reliability analysis for corresponding and future data provided in such context or improve an available monitoring device by replacing or retraining software agents providing lower quality results . For example , the skilled person can easily provide a first evaluation of the reliability of the alert by reviewing the pattern of software agents that raised the alert .

[ 0034 ] According to further embodiments it is preferred that the method contains the step of the monitoring device providing suggestions on how to operate the at least one continuous flow engine or the at least one continuous device to a user interface to be reviewed by an operator, wherein assessment data of the operator is sent from the user interface to the monitoring device , wherein the monitoring device optimi zes the software agents utili zing the assessment data . Such optimi zation of the software agents can be done using typical options available to the skilled person . For example , it was noted that typically it is especially preferred that the artificial intelligence of at least a part of the software agents is retrained, predefined rules are added, activated or modified, and/or an evaluation factor is added . Such evaluation factor speci fies the reliability of evaluations and suggestions as provided by the corresponding software agents . For example , it speci fies a software agent being less reliable resulting therein that in case of di f ferent suggestions of multiple software agents providing di f fering suggestions the software agents providing the more positive evaluation factors decide on what evaluation or suggestion is forwarded .

[ 0035 ] According to further embodiments it is preferred that the method contains the step of the monitoring device providing a suggestion on how to operate the at least one continuous flow engine and/or an assessment of a state of the at least one continuous flow engine or how to operate the at least one continuous device and/or an assessment of a state of the at least one continuous device , wherein the method contains the step of the monitoring device providing involvement data, wherein the involvement data contains data what software agents were involved with the suggestion and/or the assessment . It was noted that evaluating whether a certain suggested of operating or assessment proves to be a di f ficult task for an operator especially in case a quick reaction is required . Herein, it was noted that even after short time the skilled person can highly reliably make a first evaluation being satis fying for many to most cases based on the above speci fied accompanying data . Providing more time to spend on evaluating complex tasks or suggestions and assessments being more di f ficult to evaluate .

[ 0036 ] According to further embodiments it is preferred that at least 50% , more preferred at least 80% , even more preferred all , of the plurality of software agents have been trained using simulated data . It was noted that simulated data is a surprisingly beneficial source for such training . Especially, as the huge number of software agents can be further tuned utili zing such simulated data according to very speci fic conditions . Allowing to easily adapt the monitoring device to the very speci fic application .

[ 0037 ] According to further embodiments it is preferred that the method contains the step or retraining at least a part , preferably all , of the arti ficial intelligences of the plurality of software agents during maintenance or during utili zation, preferably during utili zation, of the at least continuous flow engine or the at least one continuous device . It should be expected that such retraining during the usage might result in an instability or di f ferent problems . However, it was noted that it is not only possible to reali ze such retraining during use . In fact , this possibility allows to very easily adapt the software agents on demand according to speci fic needs without relevant problems . In fact , it was noted that the inventive method utili zing the plurality of software agents allows to already provide a highly tailored system being very resistant to problems arising from changes . Typically, it is preferred that such retraining of the at least part , preferably all , of the arti ficial intelligences is not executed at the same time , but single or batchwise retraining is executed until all of the intended arti ficial intelligences have been retrained .

[ 0038 ] According to further aspects it is preferred that the part of the arti ficial intelligences of the aforementioned retraining is at least 10% , more preferred at least 25% , even more preferred at least 40% , of the arti ficial intelligences . It was noted that such signi ficant number of arti ficial intelligences can be retrained at a time without problem .

[ 0039 ] According to further embodiments it is preferred that the monitoring device retrieves simulation data from a simulation database , wherein the simulation database contains simulated operation data of the at least one continuous flow engine or the at least one continuous device , wherein the monitoring device retrains at least a part of the software agents based on the simulation data . Typically, it is preferred that the arti ficial intelligences are retrained during the at least one continuous flow engine or the at least one continuous device is operating . In such case , it is typically preferred to create a copy of the corresponding arti ficial intelligences , retrain them parallel to the operation of the original , and replace the original by the retrained copy . According to further embodiments it is preferred to keep the original of the arti ficial intelligence at least temporarily, like for at least 1 h, more preferred at least 24h, to allow the retrained version to instantly be replaced in case of problems .

[ 0040 ] According to further embodiments it is preferred that the at least one continuous flow engine or the at least one continuous device is operated in a fully automatic mode utili zing the monitoring device . Especially, the high reliability and possibility to easily further improve the arti ficial intelligences of the plurality of software agents by replacing them as indicated by provided evaluations and suggestions enables to provide a fully automatic mode controlling even highly complex devices like continuous flow engines or continuous devices representing highly sophisticated devices resulting in potentially catastrophic outcomes in case a grave problem occurs .

[ 0041 ] According to further embodiments it is preferred that the method contains the step of replacing at least a part , preferably all , of the arti ficial intelligences of the plurality of software agents during maintenance or during utili zation, preferably during utili zation, of the at least continuous flow engine or the at least one continuous device . It was noted that the inventive method also allows to replace a part of the arti ficial intelligences of the plurality of software agents to very easily adapt the monitoring device to corresponding changes . For example , in case of upgrades or replacements during servicing the continuous flow engines or continuous devices the arti ficial intelligences trained to monitor segregated parts containing the replaced parts can be replaced to instantly adapt the monitoring device to such change . Making the inventive method highly valuable to provide a constantly evolving and adapting system providing best results .

[ 0042 ] According to further embodiments it is preferred that the method contains the step of storing a copy of at least one of the arti ficial intelligences of the plurality of software agents . Typically, it is preferred to store such copy to review it later in case of problems . However, in most cases such copy can be beneficially utili zed as backup copy . In many embodiments it is preferred that the method contains the step of receiving a replacement command from a user interface , wherein the monitoring device replaces the arti ficial intelligences of at least a part of the plurality of software agents as referenced in the replacement command and/or based on a selection of the monitoring device based on data and/or a referenced current state as required . It was noted that in the rare case that a problem occurs such simple replacement possibility allows to instantly trigger a reset of a part or the complete plurality of software agents to fall back into a failsafe mode in case some arti ficial intelligences developed in an undesired way .

[ 0043 ] According to further embodiments it is preferred that the method contains the step of storing a copy of at least one of the arti ficial intelligences of the plurality of software agents , wherein the copy of the at least one of the arti ficial intelligences of the plurality of software agents is retrained, preferably based on simulated data, wherein the method contains the step of replacing the corresponding artificial intelligences being active in the monitoring unit by the retrained copy of the at least one of the arti ficial intelligences of the plurality of software agents , wherein the replacing takes place during maintenance or utili zation, preferably utili zation, of the at least one continuous flow engine or the at least one continuous device . Such embodiments provide an even smoother retraining process without losing the monitoring of the software agents even for a shortest time .

[ 0044 ] According to further embodiments it is preferred that the plurality of software agents generate analysis data, wherein the analysis data refer to a monitoring of the segregated parts of the industrial plant , wherein the monitoring device identi fies a failure of the at least one continuous flow engine or the at least one continuous device by exchanging analysis data between di f ferent software agents or by comparing analysis data provided by di f ferent software agents . Using the segregated feedback relating to di f ferent known parts of the device monitored allows to gain surprisingly beneficial insight by taking into account the source of the data provided . For example , allowing the software agents to exchange data allows them to directly veri fy an observation to forward a more detailed evaluation . For example , failure data provided by certain pattern of the software agents allows to deduct what causes the failure data . Without even reviewing the respective data provided itsel f . Providing a signi ficant benefit in such cases and simpli fication of further processing the acquired data despite the increased complexity of the overall system .

[ 0045 ] According to further embodiments it is preferred that the method contains assigned or changing priority data to at least a part of the plurality of software agents . Such priority data can be a simply high and low priority or any type of scale allowing to define a di f ferent priority value of such software agent . This enables an operator to very easily adapt the monitoring of the monitoring device to reflect a higher and lower quality of corresponding software agents . But especially, it was noted that a skilled person is easily able based on his/her experience to define higher and lower reliability software agents based on the assigned segregated parts . Allowing to very easily manually adapt such arti ficial intelligence-based system according to the very speci fic application case and yearlong experience . Simultaneously, this experience does not merely improve the work of di f ferent operators . Such di f ferent operators are , for example , able to make use of this available data and to gain a detailed insight when reviewing such priority data set by some experienced colleague . Preferably, the method also contains retrieving priority data and data regarding the corresponding segregated parts and forward it to a user interface to be reviewed .

[ 0046 ] According to further embodiments it is preferred that the plurality of software agents exchange data, wherein at least two , preferably at least five , even more preferred at least ten, software agents work together to identi fy a failure based on the data exchanged, wherein data regarding the failure is forwarded to the monitoring device . Typically, it is preferred that the identi fied failure is forwarded to a database like a distributed database and/or to a user interface to be reviewed by an operator . Based on the segregation of the monitoring such evaluation is surprisingly easy and for typical cases to be evaluated the evaluation can be directly reali zed in such way . Utili zing the increased insight originating from the inventive method .

[ 0047 ] According to further embodiments it is preferred that the method contains the step of automatically providing a suggestion to improve the operation of the at least one continuous flow engine or the at least one continuous device based on data provided by the plurality of software agents . Typically, it is preferred that the method contains the step of automatically implementing the provided suggestion . In this context , it is typically preferred to flexibly allow such automatic implementation of the suggestion based on a whitelist of allowed suggestions , a risk analysis of potential impact of implementing such suggestion and/or a mode to be selected .

[ 0048 ] According to further embodiments it is preferred that the method contains the step of providing a potential cause of failure in case a failure occurs within the industrial plant being monitored by the monitoring device . Besides the increase insight of the inventive method the system as described provides many possibilities to make further use of the opportunities provided herewith including directly identi fying the potential cause of a failure . One method that was very useful for typical embodiments it to identi fy such cause by comparing the pattern of observations to some historic database or simulation storing a correlation of causes of failure and corresponding patterns . Furthermore , a simulation can also be directly utili zed to do so . Herein, such pattern refers to what observations are provided by what software agents . Representing a chain of reactions resulting from such failure cause . Like a dirtied fuel starting with an increased power consumption of pumps to provide the required amount of fuel , subsequently a reduced pressure within the fuel system . Followed by an observation regarding a burning disturbance based on an incorrect mixture of fuel and air to provide the intended temperature of the burner . And many steps in between and after . Leading to a chain allowing to instantly provide such information to the operator .

[ 0049 ] For many applications it is typically preferred to speci fy the segregated parts of the industrial plant based on knowledge graph databases . According to further embodiments it is preferred that the segregated parts of the industrial plant are speci fied based on knowledge graph layers . Typically, it is preferred that the method contains adapting the segregated parts of the continuous flow engines or the continuous device during maintenance or use , preferable during use , of the continuous flow engine or the continuous device . It was noted that knowledge graph databases represent very beneficial material to provide such segregation . Especially, it was noted that the interrelation as contained in such knowledge graph database and especially the layers contained therein allows to speci fy parts of the described monitored device and interrelated components being functionally related . Allowing to speci fy a segregated parts working as a subunit and providing a speci fic behavior . Including speci fic failure modes and the like allowing to even automatically and very easily speci fy a reasonably selected segregated part . Including the possibility to automatically adapt a change reflected in an updated knowledge graph database to adapt to new, removed, or changed relations like new sensors , replacement of components by higher ef ficiency components , upgrades by including faster security elements and the like . Signi ficantly increasing the implementation and long-term usage of the inventive method over years and decades being one of the core challenges for continuous flow engines and continuous devices as mentioned herein .

[ 0050 ] According to further embodiments it is preferred that the method contains the step of the monitoring device receiving knowledge graph layers regarding the industrial plant and/or the at least one continuous flow engine or the at least one continuous device , wherein the monitoring device adapts the segregated parts of the industrial plant based on the knowledge graph layers . Typically, it is preferred to combine such step with a replacement and/or retraining of at least a part of the plurality of software agents based on the adapted segregated parts . It was noted that the inventive method allows to even implement such adaption of the segregated parts of the industrial facility during use of the at least one continuous flow engines or the at least one continuous device . Especially, as typically a complete replacement or retraining of the software agents is not required this very deep change of the utili zed system architecture is possible while the only strictly required replacement or retraining of software agents is done as required .

[ 0051 ] According to further embodiments it is preferred that the method contains a step of receiving response data from the interface , wherein the response data is utili zed by the monitoring device to evaluate the software agents . Such evaluation of the software agents typically refers to their output . For example , the suggestions provided by the software agents . Typically, it is preferred that the monitoring device marks software agents providing good and/or bad suggestions . According to further embodiments it is preferred that the monitoring device provides a list of marked software agents to a user interface and/or forwards such data to a database like a remote database . Such remote database can, for example , be located at the site of a third parts being tasked with servicing the at least one continuous flow engine or the at least one continuous device or manufacturing the at least one continuous flow engine or the at least one continuous device . It was noted that such embodiment allows to provide a surprisingly simple possibility to adapt a generic monitoring device to very speci fic needs or to very quickly identi fy the need to adapt the utili zed software agents to the very speci fic application . The modular system utili zing a plurality of software agents allows to easily replace speci fic arti ficial intelligences while the overall monitoring surprisingly not only is not impaired by such change , but even is easily improved and adapted such way . [0052] The following detailed description of the figure uses the figure to discuss illustrative embodiments, which are not to be construed as restrictive, along with the features and further advantages thereof.

[0053] Figure 1 shows a generic scheme of a system utilizing the inventive monitoring device and method. Herein, the industrial plant contains multiple continuous flow engines 2, 2' being gas turbines. However, only two of the continuous flow engines 2, 2' are shown. Herein, the continuous flow engines are a preferred and especial beneficial example, however, could also be a different continuous device as described herein also providing very beneficial results. Furthermore, the industrial plant contains a fuel supply 3 and cooling air supply 4 as shown providing fuel and cooling air to the continuous flow engine 2. The correspondingly required fuel supplies and cooling air supplies for the remaining gas turbines are not shown to simplify the figure. Also not shown is the plurality of additional segregated parts 21, 22, 23, 24, 25, 26 being monitored by the plurality of software agents 41, 42, 43, 44, 45, 46 available on the monitoring device 1. The total number of segregated parts 21, 22, 23, 24, 25, 26 and corresponding artificial intelligences 31, 32, 33, 34, 35, 36 of the software agents 41, 42, 43, 44, 45, 46 exceeds 50 per continuous flow engine 2, 2' being monitored. The corresponding segregation is based on knowledge graph layers utilized to cluster relevant subunits providing the insight to allow an artificial intelligence 31, 32, 33, 34, 35, 36 to provide relevant evaluations. While the total number typically stays the same or even increases it is possible that the segregated parts 21, 22, 23, 24, 25, 26 can be redefined. The system as shown in figure 1 allows to upload new knowledge graph layers into the monitoring device to amend the segregation resulting in, for example, additional sensor data to be included in specific parts. Herein, the number refers to the total number of all the segregated parts 21, 22, 23, 24, 25, 26 contained in the corresponding continuous flow engines 2, 2' and the respectively available auxiliary units related to the specific continuous flow engine being required for its operation.

[0054] As shown in figure 1 most segregated parts 21, 22, 23, 24, 25, 26 are not overlapping. However, for example segregated part 22 and segregated part 23 are partially overlapping representing that the provided sensor data refers to at least partially the same part of the continuous flow engine 2. In said case a sensor monitoring the overlapping part of the continuous flow engine 2 is utilized by both software agents 42 and 43 to monitor the current state. The sensors of the segregated part 22 monitor the fuel supply of the burners based on flow sensors located at the entry and the exit of the fuel supply at the burner part of the continuous flow engine 2. Simultaneously, the segregated part 23 refers to the burner temperature and utilizes not only a temperature sensor, but also the fuel supply at the exit of the fuel supply at the burner part and the air supply as well as further data to detect deviations of the temperature development not correlating to the fuel supply, air supply, and the like.

[0055] To provide correspondingly specifically adapted artificial intelligences 31, 32, 33, 34, 35, 36 more than 90% of them have been adapted using simulated data. The remaining artificial intelligences 31, 32, 33, 34, 35, 36 have been trained using historic data collected in the past. However, in case a retraining of the artificial intelligences 31, 32, 33, 34, 35, 36 is required said retraining can be based on historic data or simulated data as required for the specific situation. Herein, such retraining of the artificial intelligences 31, 32, 33, 34, 35, 36 of a part of the software agents 31, 32, 33, 34, 35, 36 can be executed on demand including during utilization of the continuous flow engines 2, 2' . Allowing to adapt the software agents 41, 42, 43, 44, 45, 46 at any time and, for example, avoiding further delays after upgrades as the corresponding adaption and fine tuning can be executed after restarting the continuous flow engines 2, 2' . Herein, the artificial intelligence 31, 32, 33, 34, 35, 36 is either retrained without further action, a copy of the artificial intelligence 31, 32, 33, 34, 35, 36 is stored while the utilized version is retrained or a copy is stored and retrained, before the retrained version replaces the original version utilized in the meantime. For example, the most important software agents being vital to a secure monitoring are retrained as copy before the original is replaced. While many artificial intelligences 31, 32, 33, 34, 35, 36 primarily providing data to support corresponding evaluations or providing a higher rate of incorrect evaluations resulting in a lower priority assigned to them are directly trained without copying it. The artificial intelligences 31, 32, 33, 34, 35, 36 of medium priority are directly trained, however, for the sake of security a copy is temporarily stored to replace them in case a problem occurs during retraining. While typically a permanent copy of the most important artificial intelligences is kept securing the availability of a failsafe version in case of problems.

[0056] Herein, sensors provide sensor data regarding the current state of the segregated parts 21, 22, 23, 24, 25, 26 to the monitoring device 1. Said sensor data is processed by the plurality of software agents 41, 42, 43, 44, 45, 46 each containing an artificial intelligence 31, 32, 33, 34, 35, 36 to each monitor the assigned segregated part 21, 22, 23, 24, 25, 26. Based on the evaluation the received sensor data by the plurality the software agents the monitoring device 1 keeps track of the current state of the continuous flow engines 2, 2' and their development. Herein, the artificial intelligences are not operating completely independent, but are additionally required to take into account predefined rules 5 that are in this case locally stored in the monitoring device 1. Said rules can, for example, contain exclusion rules that certain software agents 41, 42, 43, 44, 45, 46 are not allowed to trigger an alarm unless certain other software agents 41, 42, 43, 44, 45, 46 support such evaluation. Additionally, a ranking can be defined that specifies certain software agents 41, 42, 43, 44, 45, 46 as high priority software agents 41, 42, 43, 44, 45, 46. Such high priority software agents 41, 42,

43, 44, 45, 46 trigger an action like an alarm or a specific evaluation even in case a different software agent of lower priority provides a contrary evaluation.

[0057] Herein, the evaluation of the software agents 41, 42, 43, 44, 45, 46 include that failure modes are identified. Corresponding data regarding the failure mode is automatically assigned as label to the related data and stored in a database. Allowing to very easily create a labelled database to be utilized in the future benefitting tremendously from the high reliability of failure identification of the software agents 41, 42, 43, 44, 45, 46.

[0058] Furthermore, the evaluation data as provided by the software agents 41, 42, 43, 44, 45, 46 is provided together with evaluation data regarding an identification of a failure or a solution provided for such failure. Said evaluation data contains data how the software agents identified the failure and/or provided the solution. While the high reliability of the inventive method hardly results in conflicting or unclear evaluations of the software agents 41, 42, 43, 44, 45, 46 being required to be finally decided by a person it was noted that a skilled person like an operator receiving said data can very easily make a decision supported by this data. Corresponding feedback received in such cases or other cases requiring a feedback can be further beneficially utilized to optimize the software agents accordingly. Herein, the artificial intelligence of at least a part of the software agents is retrained, predefined rules are added, activated or modified, and/or an evaluation factor is added. Such evaluation factor specifies the reliability of evaluations and suggestions as provided by the corresponding software agents.

[0059] In case the monitoring device provides a suggestion on how to operate the at least one continuous flow engine or an assessment of a state of the continuous flow engines 2, 2' involvement data is included. Such involvement data contains data what software agents were involved with the suggestion or the assessment . It was noted that after short time such data can already be utili zed by an operator to make a first j udgement with regard to the reliability of the suggestion or assessment . Being an important improvement in case the corresponding suggestion or assessment has grave consequences when implemented like a required power reduction to avoid damages from overheating in case of oil pump mal functions or the like .

[ 0060 ] The present invention was only described in further detail for explanatory purposes . However, the invention is not to be understood being limited to these embodiments as they represent embodiments providing benefits to solve speci fic problems or ful filling speci fic needs . The scope of the protection should be understood to be only limited by the claims attached .