Designing of architecture of an intelligent management system for assets by using hydrotreating process as an example Andieva E. Yu. and Kolganov I.P. Omsk State Technical University, Pr. Mira 11, Omsk, Omsk region, Siberian federal region, Russia 55_elena@mail.ru Abstract. The article suggests the architecture solution based on the digital twin object model. The solution is based on a synthesis of relevant knowledge about international concepts of digital transformation of production systems. The design model is an information object model. The architectural concept is based on that the state of values of properties material assets, petroleum products, is affected by the state of the values of the parameters of the role equipment. On the one hand, the process control contour boundary is a cost center, on the other hand – is a profit center. The presented model is a digital twin of the hydrotreating process. The architecture solution can be replicated for other technological processes (or their segments) of production, which have continuous production in their life cycle. The architecture is consistent with the architecture of IoT, provides the ability to use synthesized machine learning technologies, takes into account landscape interoperability, provides business logic SAP ERP. Designing model approach as part of a software solution bases on the principles of system engineering (Model-based systems engineering, MBSE, system modeling notation SуSML). Keywords: Architecture Solution, Assets, Digital Twin, Hydrotreating, System Engineering, Neural Network Technology. 1 Formulation of the problem The main focus of the article is the management system for assets as a result of the business activities of a vertically integrated oil and gas company (VIOC) at the stage of refining petroleum products at the hydrotreatment process. The Design of digital twins in I4.0 solutions is a main way. There are different points of view on this question. A digital twin is a synergistic mathematical model that aggregates knowledge about an object and / or a system of objects [1]. Accept the fact: «one of the most important steps in engineering design is the choice the structure of the designed system or analysis of structural representations of the behavior and the system activity. Even, if we have a detailed mathematical model, it does not fit for this purpose» [2]. The question is one of priorities. Copyright © 2020 for this paper by its authors. Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0) ICID-2019 Conference Let’s accept that the system view of the digital twin complex manufacture control system [3, 4]. The chosen approach is consistent with Gartner agency definition of a digital twin, which based on the consolidated opinion of experts: «A digital twin is a digital representation of a real-world entity or system. The implementation of a digital twin is an encapsulated software object or model that mirrors a unique physical object, process, organization, person or other abstraction. Data from multiple digital twins can be aggregated for a composite view across a number of real-world entities, such as a power plant or a city, and their related processes» [5]. To design the digital twin for state assets management, the hydrotreating process as an example, we use a model-based methodology of system engineering – MBSE [6- 8]. This methodology has a high development rate today in practice [9-12]. SySML technology, as part of the MBSE methodology, will be used as the main design tool in the study of complex systems of various nature, which should include a hydrotreatment system [12]. 2 Design of structural view of the behavior and activities of the hydrotreatment system The concept of Digital Transformation of Petrochemical Production was proposed in [3]: «The main object of management in the new client-driven dynamic business model I4.0 is the product. As the unit of management, the state of the product is taken at a certain stage of its life cycle. This concept solves one of the key problems of the disagreement between the principles of continuous production with a multitude of simultaneous parallel production processes determined by petrochemical technologies and the discrete nature of all software systems (SS)». Particular attention is given to the digital reference architecture transformation concept [3]. As the main material asset, we denote control object – real oil state overall value chain of the production process from crude oil to marketable products of various nomenclatures – digital twin based on actually measured [4]. Assume a feature point of view: value chain – corresponds to the stages of the material asset life cycle, oil, at the production stage (provided that the oil product is a system [3, 13, 14]); production stage is a set of technological processes: process segments, operation, operation segments [15]; process boundaries determine where profit center according (SAP logic) [16]. To ensure the required actual condition of the control object, the oil product requires resources. Denote the asset systems that provide the required state of material assets at the process boundary. These are resource objects – equipment, staff and other (not related to the place of the Profit center according). Assume a feature point of view: object resource equipment at the stage of its operation (provided that the equipment is a system [3, 13, 14]); production stage is a set of technological processes: process segments, operation, operation segments [15]; process boundaries determine where cost center accounting (SAP logic) [17]. Conceptual view of technological process management system (see Fig. 1). We use the domain approach [15]. To describe structure of technological process management system was used structure diagram, which designing according SysML [18]. The conceptual view of product is represented on Fig. 2. It consists list of products from hydrotreating process [19]: diesel, hydrocarbon gas, hydrogen gas, MEA, gasoline, sour gas. Also, it shows connection (dependences) between products. The block diagram is represented on Fig. 3, which is the object model of structure. Conceptual view of equipment of the hydro treating process: ReactorBlock, StabilisationBlock, CleaningGasBlock, GasolineCleaningBlock. 1 «Subsystem» 1 «Subsystem» Product Equipment 1 «Subsystem» Process_hydrotreating 1 «Subsystem» 1 «Subsystem» Others People Fig. 1. Conceptual view of technological process management system (Structure diagram technological process management system) 1 «Subsystem» Product 1 «Subsystem» 1 «Subsystem» 1 «Subsystem» Diesel Hydrogen_gas Hydrocarbon_gas 1 «Subsystem» 1 «Subsystem» 1 «Subsystem» Gasoline MEA Sour_gas Fig. 2. Conceptual view of oil product subsystem (Structure diagram oil product subsystem) Assume a feature point of view: accept the fact that the equipment is divided into equipment and role equipment [4, 15]; the state of the values of the parameters of the role equipment affect the state of the values of the properties of the material asset, oil product. 1 «Subsys tem» Equipment 1 «Subsys tem» 1 «Subsys tem» ReactorBlock StabilisationBlock 1 «Subsys tem» 1 «Subsys tem» CleaningGasBlock GasolineCleaningBlock Fig. 3. Conceptual view of equipment subsystem (Structure diagram equipment subsystem) Conceptual view of subsystem equipment gasoline cleaning block is shown on Fig. 4. Assume a feature point of equipment view: the equipment is divided into segments according to technological process. Denote the block for moving the monoethanolamine to the other block – MEAPumpInBlock, which include role equipment: MEA transferring pumps: Н_5_1, Н_5_2. Denote the unit for extraction process – ExtractorBlock that consist extractor K_6 и tank Е_6. Denote the block for moving the monoethanolamine from the other block – MEAPumpOutBlock: MEA transferring pumps: H_6_1, H_6_2; Denote the block for moving the gasoline out – GasolinePumpOutBlock: gasoline pumps H_3_1, H_8_1; Denote the gasoline production block – GasolineProductionBlock – the source of raw material – L_24_6, L_24_7. GasolineProductionBlock is an outside system of equipment, which transfer gasoline for cleaning technological process as a one of the material asset. Conceptual view of subsystem equipment stabilization block is shown on Fig. 5. Assume a feature point of equipment view: the equipment is divided into segments according to technological process. Stabilization block is the next stage, several technological processes, after Reactor block, which include hydrogenation process – purification of diesel fuel from harmful impurities oxygen, sulfur, nitrogen, by using hydrogen and catalysts, reactions occurring under certain technological state. Separated hydrocarbons are sent to stabilization block. According to technological operations, the equipment included in this unit divided into groups: StabilisationColumnBlock, SeparationBlock_of_StabilisationBlock, GasolineOutPumpBlock, DieselOutCoilsBlock. Marked block for stabilization process of diesel fuel – StabilisationColumnBlock. It consists stabilization column K_1 and related equipment – irrigation pumps H_3_1, H_3_2; kilns for heating diesel in the boiler P_2_R, P_2_L; diesel pumps for movement part of the fuel to the boiler and other part out H_2_1, H_2_2, H_2_3. Next block is needed for separate substance, one of them direct to K_1, part of them go out, and gas move to the other technological block. Denote the separation block, which divide substance into several layer – SeparationBlock_of_StabilisationBloc. In include some technological equipment: for substance cooling two air coils XK_1_1, XK_1_2, and one liquid coil – X_10, and separator C_5; Denote the block, which responsible for the withdrawal of stable gasoline from process – GasolineOutPumpBlock. This block consists two diesel pumps H_3_3, H_8_1. Denote heat exchange block – Dieseloutcoilsblock. This group of equipment is a place of interaction with other blocks – Reactor block and cleaning gas block. There are exchangers for MEA – T_4; diesel exchangers T_2_1 _n, T_2_2_n, T_2_3_n, T_2_1_v, T_2_2_v, T_2_3_v; exchangers between new diesel and diesel after hydrotreating T_26, T_27; and the last air cooler for diesel X_5_1, X_5_2, X_5_3. Assume a feature point of view: equipment – an object of management; equipment can perform certain technological functions – it is role-playing equipment. To make managerial decisions, we will consider only those parameters that affect the properties of the main material asset, oil product. Consider the properties of role-based equipment using object – compressor PK_1_1 as an example. First, we highlight the technological parameters: temperature and pressure in the compressor row. For a material asset, diesel and gasoline, it is necessary to maintain the required properties, which are achieved by the required chemical composition of the substance under certain environmental parameters, which are provided by a certain value of the role equipment parameters. Therefore, we identified two main parameters of the compressor. To maintain normal parameters, it is necessary to maintain a certain, operational condition of the equipment. The condition of the equipment greatly affects its operation, and as a result, the main material asset. The compressor is a dynamic equipment. Dynamic properties of compressor: vibro movement, vibro speed, vibro acceleration. The listed characteristics are measured on the crankshaft of the compressor. The rotation of the shaft and its vibrations are affected by bearings. The condition of the bearings can be monitored indirectly by measuring their temperature. Therefore, six more parameters are added according to the number of bearings. It is possible to control the compressor through a system that squeezes the suction valve when compressing gas. Therefore, the parameters in this row will change differently, which also affects the output state of the properties of the material asset. So, the state of the values of the parameters of the role equipment affect the state of the values of the properties of the material asset, oil product. So, compressor like role equipment has some parameters: activity – one, two, three or four line, bearings temperature: first bearing, second bearing, third bearing, fourth bearing, fifth bearing, sixth bearing; line parameters (temperature, pressure), vibro acceleration, vibro movement, vibro speed. On the next stage, parameters are divided from equipment. For example, get object – compressor PK_1_1. We can see the object, which describe parameters of the compressor on the Fig. 6. Each object is a specific compressors parameter. Let’s divide substance properties and parameters of the equipment that we will have a properties layer. So, we can see equipment parameters of the compressor. This layer will be used in the neural network. 1 «Subsyst em» 1 H_3_3:Pump L_24_6:Production be a ringTe mpe ra ture [1.. .. . id:RhpString 1 GasolinePumpOutBlock 1 na me :RhpString GasolineProductionBlock product ion:double roundSpe e d:double subst a nce In:Subst a nce subst a nce O ut:Substa nce 1 «Subsyst em» 1 «Subsyst em» cha nge RoundSpe e d(rou.. . L_24_7:Production Equipment::GasolineCleaningBlock compre ss(substa nce In:S. .. 1 MEAPumpInBlock ge t Subst a nce O ut():Subs.. . of f():void on():void 1 ExtractorBlock 1 MEAPumpOutBlock 1 H_5_1:Pump 1 H_5_2:Pump 1 H_8_1:Pump be a ringTe mpe ra ture [1. . *]:double be a ringTe mpe ra ture [1.. *]:double be a ringTe mpe ra ture [1. . *]: . .. id:RhpString id:RhpString id: RhpString na me :RhpString na me :RhpString 1 1 H_6_1:Pump 1 H_6_2:Pump na me : RhpString K_6:Extractor product ion:double product ion:double 1 E_6:Tank product ion: double be a ringTe mpe ra ture [1. . *]:. .. be a ringTe mpe ra ture [1.. .. . roundSpe e d:double roundSpe e d:double subst a cne He a vyO ut :. .. roundSpe e d: double volume:do.. . id:RhpString id:RhpString subst a nce In:Subst a nce subst a nce In:Subst a nce subst a nce He a vyIn: S. .. subst a nce In: Subst a nce na me :RhpString na me :RhpString subst a nce O ut:Substa nce subst a nce O ut:Substa nce subst a nce Light In:Su. .. subst a nce O ut: Subst a nce product ion:double product ion:double subst a nce Light O ut : .. . cha ngeRoundSpe e d(roundSpe e d:i. .. cha nge RoundSpe e d(roundSpe e d. . . roundSpe e d:double roundSpe e d:double cha ngeRoundSpe e d(round. . . compress(substa nce In:Substa nce ):. .. compre ss(substa nce In:Substa nce . .. subst a nce In:Subst a nce subst a nce In:Subst a nce compress(substa nce In:Subs.. . ge t Subst a nce O ut():Substa nce ge t Subst a nce O ut():Substa nce subst a nce O ut:Subst a nce subst a nce O ut:Substa nce ge t Subst a nce O ut():Substa nce of f():void of f():void of f():void on():void cha ngeRoundSpe e d(round. . . cha nge RoundSpe e d(rou.. . on():void on():void compress(substa nce In:Subs.. . compre ss(substa nce In:S. .. ge t Subst a nce O ut():Substa nce ge t Subst a nce O ut():Subs.. . of f():void of f():void on():void on():void Fig. 4. Conceptual view of gasoline cleaning block (Block diagram subsystem equipment gasoline cleaning block) 1 X_5_1:AirCoil 1 X_5_3:AirCoil id:RhpString 1 T_2_2_v:Exchanger 1 T_26:Exchanger name:RhpStri ng id:RhpString 1 X_5_2:AirCoil 1 T_2_1_v:Exchanger substanceColdI n:Su... open:bool name:RhpString substanceColdI n:Su... substanceColdO ut:S... id:RhpString roundSpeed:i nt substance ColdI n:Substance open:bool T_27:Exchanger substanceColdO ut:S... name:RhpString s ubs tanceH otI n:S... 1 roundSpeed:int substance ColdO ut:Substance open:bool s ubs tanceH otO ut:... exchange( hotI n:Sub... substanceH otI n:S... substanceCol dI n:S... roundSpeed:int tEnv:double exchange( hotI n:Sub... getColdO ut( ) :Substa... substanceH otO ut:... substanceCol dO ut:... exchange( hotI n:Substance... substanceH otI n:S... getColdO ut( ) :Substa... getH otO ut( ) :void tEnv:double getH otOut( ) :void getColdO ut( ) :Substance substanceH otO ut:... getH otO ut( ) :void getH otO ut( ) :void tEnv:double exchange( hotI n:S... getH otO ut( ) :void 1 T_2_3_n:Exchanger getColdO ut( ) :Subs... getH otO ut( ) :void getH otO ut( ) :void substance ColdI n:Substance 1 DieselOutCoilsBlock substance ColdO ut:Substan... 1 T_2_2_n:Exchanger 1 H_3_2:Pump 1 K_1:StabilisationColumn substance ColdI n:Substance exchange( hotI n:Substanc... bearingTempera... substance ColdO ut:Substance getColdO ut( ) :Substance id:RhpString heavySubstanceI n:Substance 1 T_2_1_n:Exchanger getH otO ut( ) :void 1 T_4:Exchanger name:RhpString lightSubstanceI n:Substance exchange( hotI n:Substance... production:double substanceCol dI n:S... substanceColdI n:Sub... getColdO ut( ) :Substance roundSpeed:dou... substanceCol dO ut:... substanceColdO ut:Su... getH otO ut( ) :void 1 T_2_3_v:Exchanger s ubs tanceI n:Su... 1 «Subsystem» exchange( hotI n:S... s ubs tanceO ut:S... Equipment::StabilisationBlock getColdOut( ) :Subs... exchange( hotI n:Subs... substance ColdI n:Substance getColdO ut( ) :Substan... substance ColdO ut:Substance changeRoundSp... getH otO ut( ) :void compress( s ubs... 1 StabilisationColumnBlock exchange( hotI n:Substance... getSubstanc eO ... getColdO ut( ) :Substance off( ) :void 1 SeparationBlock_of_StabilisationBloc k 1 GasolineOutPumpBloc k getH otO ut( ) :void on( ) :void 1 P_2_R:Kiln airProductio... 1 H_2_1:Pump fireTemperat... substanceI ... 1 H_3_3:Pump 1 H_8_1:Pump bearingTempera... 1 H_2_2:Pump 1 H_2_3:Pump 1 P_2_L:Kiln substanceO ... 1 X_10:LiquidCoil bearingTempera... bearingTempera... id:RhpString bearingTempera... bearingTempera... 1 XK_1_2:AirCoil name:RhpString fireTemperat... heat( substa... id:RhpString 1 C_5:Separator id:RhpString id:RhpString production:double id:RhpString id:RhpString s ubs tanceI ... id:RhpString name:RhpString name:RhpString name:RhpString roundSpeed:dou... name:RhpString name:RhpString s ubs tanceO ... name:RhpString pressure:double production:double production:double substanceColdI n:Sub... substanceI n:Su... production:double production:double 1 XK_1_1:AirCoil open:bool substanceI n[ 1..*] ... roundSpeed:dou... roundSpeed:dou... substanceColdO ut:Su... substanceO ut:S... roundSpeed:dou... roundSpeed:dou... roundSpeed:int substanceO ut[ 1..*... substanceI n:Su... substanceI n:Su... id:RhpString substanceH otI n:Sub... substanceI n:Su... substanceI n:Su... heat( substa... substanceO ut:S... substanceO ut:S... substanceH otI n:Sub... substanceH otO ut:do... volume:doubl e changeRoundSp... name:RhpString substanceO ut:S... substanceO ut:S... substanceH otO ut:do... open:bool tEnviroment:double compres s ( s ubs ... changeRoundSp... changeRoundSp... tEnv:double getSubs tanc eO ... changeRoundSp... changeRoundSp... roundSpeed:int compress( s ubs... compress( s ubs... getColdO ut( ) :Substa... off( ) :void compress( s ubs... compress( s ubs... substanceH otI n:S... getSubstanc eO ... getSubstanc eO ... getH otO ut( ) :void getH otO ut( ) :Substan... on( ) :void getSubstanc eO ... getSubstanc eO ... substanceH otO ut:... off( ) :void off( ) :void refrigerate( coldI n:Su... off( ) :void off( ) :void tEnv:double on( ) :void on( ) :void on( ) :void on( ) :void getH otO ut( ) :void Fig. 5. Conceptual view of stabilization block (Block diagram subsystem equipment stabilization block) 1 temperature_bearing_1_PK_1_1 1 bearing_1_PK_1_1 1 activityPK_1_1 1 PK_1_1:Compressor 1 production_PK_1_1 1 bearing_2_PK_1_1 activity:d ouble bearingTemperatu re[1. . *].. . id :Rh pString 1 line_4_PK_1_1 1 temperature_bearing_2_PK_1_1 lineParameter[1. . *]:S ubst. . . name:RhpString 1 bearing_3_PK_1_1 1 bearings_PK_1_1 production:dou ble 1 lineParameters_PK_1_1 vibroAcceleratio n:d ouble vibroM ovement:double 1 temperature_line_4_PK_1_1 vibroSpe ed:double 1 temperature_bearing_3_PK_1_1 1 bearing_4_PK_1_1 changeA ctivity(numberActiv. . . comp ress(su bstanceIn:Subs. . . 1 pressure_line_4_PK_1_1 getSubst anceO ut():Subst a.. . 1 bearing_5_PK_1_1 1 bearing_6_PK_1_1 1 line_2_PK_1_1 1 line_3_PK_1_1 of f ():void O n():void 1 temperature_bearing_4_PK_1_1 1 temperature_line_3_PK_1_1 1 line_1_PK_1_1 1 1 1 crankshaft_PK_1_1 temperature_bearing_5_PK_1_1 temperature_bearing_6_PK_1_1 1 pressure_line_3_PK_1_1 1 vibroMovement_PK_1_1 1 vibroSpeed_PK_1_1 1 temperature_line_1_PK_1_1 1 temperature_line_2_PK_1_1 1 vibroAcceleration_PK_1_1 1 pressure_line_1_PK_1_1 1 pressure_line_2_PK_1_1 Fig. 6. Conceptual architecture of compressors parameters Take hydrogen gas as an example, because it interacts with the PK_1_1 compressor – resource object, which described above. We distinguish the following properties of the substance destiny, fair temperature, sour percent (special quality, that not included in the physicochemical properties of substances, but isolated and important) boiling temperature, autoignition temperature, ignition temperature, flash point, heat capacity. The achievement of the required properties values of substances described earlier is the goal of controlling the values of equipment parameters. To identify complex dependencies between the values of the equipment parameters, – the resource object and the material asset, – oil property values, consequently, neural network technologies will be used to calculate the control action. Neural networks are widely used for decision-making based on the analysis using technology data mining [20, 21]. The equipment parameters obtained from sensors and the state of oil products provide to the input layer of the NN. We use a multilayer neural network to identify complex hidden relationships between input and output vectors. At the output, we get a vector that describes the state of the oil product – the main asset, based on which, through communications, we will determine the most influential equipment and manage it. The target system is divided into five stage. The data from the sensors, that describes real equipment, must be collected and processed. The first component will collect data from sensors. The second component should sort into two parts: negative and positive example. Then several data problems should be solved. Firstly, some parameters or properties are analyzed once during a day or week. Second, data dimensions are various. So, temperature is measured between minus 70 and plus 700 oC, but pressure is measured between 0,1 and 6 MPa. It is a problem for NN training. Management system – is a deep neural network, which has input vector and output vector. Input vector is a data, which describes real substance property, property, that we wish, and the ideal property. Output vector is a data, which describes parameters of the equipment; owing to we can reach necessary property. 3 Summarize The main distinctive features of the proposed concept of the digital transformation reference architecture are the following. The article presents the architecture solution based on the digital twin object model. The architecture solution can be replicated for other technological processes (or their segments) of production, which have continuous production in their life cycle. The architecture is consistent with the architecture of IoT, provides the ability to use synthesized machine learning technologies, takes into account landscape interoperability, provides business logic ERP SAP. Designing model approach as part of a software solution bases on the principles of system engineering (MBSE, SуSML). References 1. Borovkov, A.I.: Digital twins and the digital transformation of defense enterprises. Bulletin of the east siberian open academy (32), 2 (2019). 2. Sadovskiy, V.N., Yudin, E.G.: Research on General systems theory. In: Makarov, A.A. Progress, Moscow (1969). 3. Andieva E. Yu., Kapelyuhovskaya A.A. New approaches to digital transformation of petrochemical production // AIP Conference Proceedings vol. «Oil and Gas Engineering, OGE 2017» 2017. С. 020076. 4. Andieva E.Y., Tolkacheva E.V. Development of methodological basis i4.0 for production systems of oil refining industry. Oil and Gas Studies. 2018; (4):137-147. (In Russ.) https://doi.org/10.31660/0445-0108-2018-4-137-147. 5. Gartner Glossary. https://www.gartner.com/en/information- technology/glossary/digital-twin, last accessed 2019/12/10. 6. Baker, L.: Foundational Concepts for Model Driven System Design. INCOSE Model Driven Design Interest Group, INCOSE, Jul. 2000. 7. Hoffmann, H.: SysML-based systems engineering using a model-driven development approach. IBM Systems Journal 25(3), 569–585 (2008). 8. Estefan, J. A.: Survey of model-based systems engineering (MBSE) methodologies. INCOSE Technical Document. Revision B, June 10, 2008. 9. Nolan, B., Brown, B., Balmelli, L.: Model Driven Systems Development with Rational Products. 1st edn. 2008. 10. Madni, A.: Leveraging Digital Twin Technology in Model-Based Systems Engineering. Systems 7(7), 1–13 (2019). 11. Schumacher, D.: From Document Centric Approach to MBSE Approach: BPMN, UML, SysML and Wire Framing Implementation. In: Bonjour, E., Krob, D. (eds) Complex Systems Design & Management. CSD&M 2018. Springer, Cham (2019). 12. Wolny, S., Mazak, A.: Thirteen years of SysML: a systematic mapping study. Software & Systems Modeling. 1–59 (2019). 13. ISO 55000, Asset management — Overview, principles and terminology, 2014. 14. IEEE 15288, Information technology. System engineering. System life cycle processes, 2015. 15. IEC 62264-1, Enterprise-control system integration — Part 1: Models and terminology, 2014. 16. Profit Center Accounting (EC-PCA), https://help.sap.com/viewer/1b28652e774d490db5310fd6e426691f/6.00.31/en- US/7765cf536db84408e10000000a174cb4.html, last accessed 2019/12/17. 17. Cost Center Accounting (CO-OM-CCA), https://help.sap.com/doc/saphelp_afs64/6.4/ru- RU/08/513e4243b511d182b30000e829fbfe/frameset.htm, last accessed 2019/12/17. 18. The OMG system modeling language specification version 1.6, https://www.omg.org/spec/SysML/, last accessed 2019/12/17. 19. Ayshten, V. G., Nosov, G. A., Zaharov, M. K.: Processes and apparatus of chemical technology. Lan, Moscow (2019). 20. Schmidhuber, J.: Deep learning in neural networks: An overview. Neural Networks. 85–117 (2015). 21. Musci, M.: A scalable multi-signal approach for the parallelization of self- organizing neural networks. Neural Networks. 108–117 (2019).