Safe operation of chemical plants

Abstract

A method for controlling and / or monitoring a target chemical plant, the method comprising: providing, in particular receiving, malfunctional data associated with a difference between - a current operation of a target chemical plant and a target operation of the target chemical plant and / or - a current configuration of the target chemical plant and a target configuration of the target chemical plant, receiving a plurality of historical plant data sets associated with a plurality of differences between - historical operations and target operations of a plurality of chemical plants and / or - historical configurations and target configurations of a plurality of chemical plants and a plurality of causes related to the plurality of differences and / or a plurality of measures for eliminating the differences, selecting at least a part of the historical plant data sets by matching at least a part of the malfunctional data and the plurality of historical plant data sets, selecting a subgroup of at least the selected part of the historical plant data sets based on a digital representation of at least the part of the historical plant data sets and a digital representation of the malfunctional data, providing the one or more measure(s) and / or one or more cause(s) associated with the selected subgroup of the historical plant data sets for controlling and / or monitoring the operation and / or the configuration of the target chemical plant.

Description

[0001] 240436

[0002] SAFE OPERATION OF CHEMICAL PLANTS

[0003] TECHNICAL FIELD

[0004] The disclosure relates to a safe operation of chemical plants and to a method for controlling and / or monitoring a target chemical plant, use of one or more measure(s) and / or cause(s) as obtained by any one the methods as described herein for controlling and / or monitoring one or more target chemical plant(s), a non-transitory computer-readable data medium, an ratus, in particular for controlling and / or monitoring one or more target chemical plant(s).

[0005] TECHNICAL BACKGROUND

[0006] Chemical plants have to be operated according to strict processes developed over years to ensure the high safety standards are met. Said processes are developed from experience by workers of the chemical plants. It is desired to improve processes even further with time to allow for the highest safety standards possible.

[0007] SUMMARY

[0008] In an aspect, this disclosure relates to a method for controlling and / or monitoring a target chemical plant, the method comprising:

[0009] obtaining, in particular receiving, malfunctional data associated with a difference between a current operation and / or configuration associated with a target chemical plant and a target operation and / or configuration associated with the target chemical plant,

[0010] obtaining, in particular receiving, a plurality of historical plant data sets associated with a plurality of differences between historical operations and / or configurations and target operations and / or configurations associated with a plurality of chemical plants and a plurality of causes related to the plurality of differences and / or a plurality of measures for reducing an effect of the differences onto the operation of the target chemical plant,

[0011] selecting at least a part of the historical plant data sets by matching at least a part of the malfunctional data and the plurality of historical plant data sets,

[0012] providing one or more digital representation(s) of at least the selected part of the historical plant data sets and a digital representation of the malfunctional data,

[0013] selecting a subgroup of at least the selected part of the historical plant data sets based on, in particular by matching, the one or more digital representation(s) of at least the part of the historical plant data sets and the digital representation of the malfunctional data,240436

[0014] 2

[0015] providing the one or more measure(s) and / or one or more cause(s) associated with the selected subgroup of the historical plant data sets for controlling and / or monitoring the operation and / or the configuration of the target chemical plant.

[0016] In another aspect, it relates to use of one or more measure(s) and / or cause(s) as obtained by any one the methods as described herein for controlling and / or monitoring one or more target chemical plant(s).

[0017] In another aspect, it relates to an apparatus, in particular for controlling and / or monitoring one or more target chemical plant(s), the apparatus comprising:

[0018] a processor configured for performing any one of the methods as described herein.

[0019] In another aspect, it relates to a non-transitory computer-readable data medium storing a computer program including instructions for executing steps of the method as described herein.

[0020] EMBODIMENTS

[0021] Any disclosure, embodiments and examples described herein relate to the methods, the systems, apparatuses, chemical products and computer elements lined out above and below. Advantageously, the benefits provided by any of the embodiments and examples equally apply to all other embodiments and examples.

[0022] In the following, terminology as used herein and / or the technical field of the present disclosure will be outlined by ways of examples. Where examples are given, it is to be understood that the present disclosure is not limited to said examples.

[0023] These and other objects, which become apparent upon reading the following description, are solved by the subject matters of the independent claims. The dependent claims refer to embodiments of the disclosure.

[0024] Chemical plants have high safety requirements to ensure error-free production as errors in the operation provide hazardous risks to workers, equipment and environment. Followingly, operation and configuration of chemical plants are checked thoroughly in highly complex and time-consuming processes. This includes retrospectively analyzing previous incidents by human experts with a lot of experience in operating plants. Documentation of such previous incidents may comprise a plurality of different data points, in particular long texts. Hence, it is desired to improve using knowledge gained in relation to historical incidents better.

[0025] This can be achieved by robustly retrieving related incidents and corresponding measures and causes to improve controlling and / or monitoring of chemical plants. In particular, retrieving related incidents objectively by matching240436

[0026] 3

[0027] digital representations allows to retrieve the relevant results in a scalable yet accurate manner. Furthermore, boundaries while evaluating texts including synonyms of target keywords are removed and search reliability is increased. By selecting a subgroup of at least the selected part of the historical plant data sets based on digital representations of a selected part of the historical plant data sets and the malfunctional data a second and more sensitive filtering of available historical plant data sets is conducted after a first, typically rougher, search. As a consequence, computational resources that would be needed to apply the sensitive second filtering of a high amount of data can be saved and implementation of a scalable application making use of the subject-matter of this disclosure is enabled. This comes along with the benefit of allowing for the globally harmonized application of safety measures independent of the experience of a human expert. Furthermore, the data to be processed by the implementation of the subject-matter of this disclosure can be human-interpretable. Thus, allowing for review by human users. In addition, the selected measures and / or causes can be provided directly to workers of chemical plants to implement solutions in order to avoid similar incidents. Ultimately, this saves time when implementing solutions which can increase resource efficiency when handling incidents and allows for responsible use of computational resources.

[0028] In an embodiment, the malfunctional data may be associated with a difference between a current operation of the target chemical plant and a target operation of the target chemical plant. Additionally or alternatively, the malfunctional data may be associated with a difference between a current configuration of the target chemical plant and a target configuration of the target chemical plant. In particular, the malfunctional data may be indicative of and / or may include the difference between a current operation and / or configuration of the target chemical plant and a target operation and / or configuration of the target chemical plant. The malfunctional data may be related to an atypical operation and / or configuration of the target chemical plant. Preferably, the malfunctional data may include an incident report and / or may be related to an incident occurred in the target chemical plant, in particular while operating the target chemical plant according to the current operation and / or wherein the target chemical plant may be configured according to the current configuration. Configuration of chemical plants may be related to an arrangement of equipment associated with the target chemical plant. Operation of chemical plants may be related to setting one or more operating parameter(s) associated with at least parts of the equipment for producing chemical products by the chemical plants. Adapting the operating parameters may result in adapting operating of at least a part of the equipment. The current operation may refer to an operation of the target chemical plant associated with an incident and / or resulting in the incident related to the target chemical plant. The current operation may refer to an operation of the target chemical plant for producing chemical products. Historical operation may refer to an operating previous to the current operation. The current configuration may refer to a configuration of the target chemical plant for producing chemical products. Historical configuration may refer to a configuration of a chemical plant prior to the current configuration. The current configuration and the historical configuration and / or the current operation and the historical operation can match and / or be different.240436

[0029] 4

[0030] In an embodiment, a cause may be related to a difference between target operation and / or configuration and current operation and / or configuration. In particular the cause may lead to and / or may result in the difference between target operation and / or configuration and current operation and / or configuration. The cause may include a root cause and / or one or more intermediate cause(s). The root cause may refer to an event causing one or more following event(s), in particular the root cause may result in and / or may lead to one or more intermediate cause(s). Removing and / or elimination the root cause may result in removing and / or eliminating the one or more intermediate cause(s). The root cause may be associated with an earlier point in time than the one or more intermediate cause(s). In an embodiment, the malfunctional data may comprise and / or may be indicative of one or more intermediate cause(s) and the plurality of causes associated with the historical plant data sets may comprise and / or may be a plurality of root causes. Additionally, the historical plant data sets may comprise one or more intermediate causes associated with the plurality of root causes. The intermediate cause may be a result of the plurality of root causes. By providing root causes, incidents can be circumvented from the beginning and measures relating to the root cause can be implemented. This improves the efficiency of the safety work related to chemical plants. Ultimately, this reduces resources used for production and therefore, improves production efficiency.

[0031] In an embodiment, a measure may be suitable for and / or configured to reduce an effect of one or more differences of the operation of the target chemical plant. In particular, the measure may be suitable for and / or configured to reduce the effect of the one or more cause(s) on the operation of the target chemical plant, in particular eliminate the one or more root cause(s) associated with the malfunctional data. Applying a measure may result in adapting the operation and / or the configuration of chemical plants, in particular the target chemical plant. A measure may be a corrective action, in particular for reducing an effect of one or more differences of the operation of the target chemical plant.

[0032] In an embodiment, the digital representation may be a computer-readable representation. The digital representation may be a numerical representation. Hence, the digital representation may comprise one or more numerical value(s) related to, in particular obtained from the malfunctional data and / or the historical plant data sets. The digital representation may be a structured representation of the malfunctional data and / or at least a part of the historical plant data sets. The digital representation may be obtained by mapping the historical plant data sets and / or the malfunctional data to a digital representation of the historical plant data set(s) and / or of the malfunctional data, in particular in a representation space. For example, the digital representation may be a tensor and / or a numerical representation of the historical plant data sets and / or the malfunctional data.

[0033] In an embodiment, providing the malfunctional data may be triggered by detecting a difference between the current operation and / or configuration and the target operation and / or configuration of the target chemical plant by one or more sensor(s) associated with the target chemical plant. The one or more sensor(s) may be configured to240436

[0034] 5

[0035] monitor operation and / or the configuration of the target chemical plant. Detecting the difference may include obtaining data related to the operation and / or the configuration of the target chemical plant by the one or more sensor(s) and comparing the obtained data with target values associated with a target operation and / or target configuration of the target chemical plant. By doing so, deviating operation and / or configuration of the target chemical plant can be obtained directly while monitoring and used for eliminating causes for potential incidents and / or decrease the time between incidents and resolving issues that lead to the incident. Thereby, production and plant resources as well as time can be saved while increasing safety for workers of a chemical plant.

[0036] In an embodiment, the one or more measure(s) and / or the one or more cause(s) may be provided via a user interface and / or displayed for implementing the one or more measure(s) and / or reducing the effect of the one or more cause(s) on the operation of the target chemical plant. The user interface may be accessible by workers associated with the target chemical plant, preferably a plurality of chemical plants. Preferably, the one or more measure(s) and / or cause(s) associated with the target chemical plant may be accessible via the user interface and / or displayed to workers associated with the target chemical plant. The one or more measure(s) and / or the one or more cause(s) may be displayed and / or provided to be implemented by one or more worker(s) associated with the target chemical plant. By doing so, measures and / or causes are directly accessible to workers for implementation. This decreases a time gap between reporting on an incident and resolving causes related to said incident.

[0037] In an embodiment, the malfunctional data and / or the one or more measure(s) and / or the one or more cause(s) may be human-interpretable and / or may comprise natural language. The malfunctional data and / or the one or more measure(s) and / or the one or more cause(s) may comprise string data and / or text data. By doing so, the malfunctional data may be provided by and / or to workers associated with the target chemical plant. This reduces a time gap between detecting a difference by workers and reporting on the difference as well as a time gap between reporting on an incident and receiving instructions for handling the incident.

[0038] In an embodiment, the malfunctional data may include sensor data collected by one or more sensor(s) for monitoring the target chemical plant. By including sensor data, objectively obtained data related to the operation and / or the configuration can be used for determining causes and / or measures associated with the target chemical plant. Furthermore, this feature allows to make use of readily available data to improve controlling and / or monitoring of chemical plants.

[0039] In an embodiment, providing the one or more measure(s) and / or cause(s) may include providing the selected subgroup of the historical plant data sets. The historical plant data sets may further comprise data related to operation and / or configuration of the chemical plants independent of the plurality of differences and / or related to a target operation and / or configuration of the target chemical plant. Optionally, a verification of the selection of the240436

[0040] 6

[0041] subgroup of historical plant data sets may be received, e.g. via a user interface, and / or an adapted selection of the subgroup of historical plant data sets may be received, e.g. via a user interface. By doing so, the selection of the historical plant data sets can be evaluated by a human expert to improve verify reliability of the selection. Thereby, it can be ensured that correct information for controlling and / or monitoring the target chemical plant is used. Further, additional information relevant e.g. for workers associated with the target chemical plant can be taken into account when implementing the measures and / or eliminating the causes.

[0042] In an embodiment, the one or more cause(s) provided may include a root cause and one or more related intermediate cause(s). The one or more intermediate cause(s) and the root cause may be part of a chain of events and / or causes related to operation and / or configuration of the target chemical plant. By doing so, further data suitable for preparing working behaviors in relation to incidents associated with the malfunctional data can be established according to the follow-up events. This reduces time to react upon a chain of events and thus, improves reliable handling of incidents by workers and / or machinery.

[0043] In an embodiment, selecting at least a part of the historical plant data sets by matching at least a part of the malfunctional data and the plurality of historical plant data sets may comprise selecting at least a part of the historical plant data sets by matching a plant type associated with the target chemical plant and plant types associated with the plurality of chemical plants. Events occurring in a chemical plant may be specific to the plant type. E.g. plant type can be a polymerization plant or a water-based reaction plant. Hence, a preselection according to the plant type allows to eliminate non-applicable measures and causes by an easy and efficient filtering.

[0044] In an embodiment, selecting at least a part of the historical plant data sets by matching at least a part of the malfunctional data and the plurality of historical plant data sets may comprise selecting at least a part of the historical plant data sets by matching one or more element(s) associated with at least a part of the malfunctional data and one or more element(s) per of historical plant data set. The one or more element(s) may comprise at least a part of a word, at least a part of a table, a numerical value and / or a symbol. Index search may be an efficient and robust filtering for data sets according to search criteria. The one or more element(s) associated with at least the part of the malfunctional data may comprise the difference between the current operation and / or configuration and the target operation and / or configuration. The one or more element(s) associated with the historical plant data sets may comprise a subsection of the historical plant data sets. The subsection of the historical plant data sets may be associated with the one or more difference(s) between the historical operation and / or configuration and the target operation and / or configuration. Additionally or alternatively, the subsection of the historical plant data sets may be associated with the one or more measure(s) and / or cause(s).240436

[0045] 7

[0046] In an embodiment, selecting the subgroup of at least the selected part of the historical plant data sets based on a digital representation of at least the part of the historical plant data sets and a digital representation of the malfunctional data may include matching the digital representation of at least the part of the historical plant data sets and the digital representation of the malfunctional data. Matching digital representations may include determining a distance associated with pairs of digital representations of malfunctional data and historical plant data sets. By doing so, a context-aware and thus, highly accurate search is used as a second search to refine the first selection.

[0047] In an embodiment, selecting the subgroup of at least the selected part of the historical plant data sets based on a digital representation of at least the part of the historical plant data sets and a digital representation of the malfunctional data may include processing of the digital representations by a data-driven model. The data-driven model may be suitable for and / or may be configured to match(ing) the historical plant data sets and the malfunctional data based on the digital representations of the malfunctional data and the historical plant data sets. The data-driven model may be configured to obtain at least one digital representation per historical plant data set and the digital representation of the malfunctional data for matching the at least one digital representation per historical plant and the digital representation of the malfunctional data. Preferably, the data-driven model may be configured to follow task instructions and selecting the subgroup of at least the selected part of the historical plant data sets may comprise providing a matching task instruction for matching the malfunctional data with the selected part of the historical plant data sets to the data-driven model. The matching task instruction may be associated with the digital representation of the historical plant data sets and the digital representation of the malfunctional data. In particular, the matching task instruction may comprise natural language and providing the matching task instruction may comprise and / or may result in mapping the malfunctional data and the historical plant data sets to the digital representation of the malfunctional data and the digital representation of the historical plant data sets. In an embodiment, the data-driven model may be a pretrained data-driven model(s), wherein the pretrained data-driven model is configured to perform a plurality of different tasks according to a plurality of different task instructions. The pretrained data-driven model may be trained and / or parametrized based on unstructured data, in particular text data and optionally numerical data such as tabular data or image data. The unstructured data may be associated with a plurality of different tasks and / or task instructions. The pretrained data-driven model(s) may be configured to perform a plurality of task. The pretrained data-driven model(s) may be configured to and / or may be suitable to perform the task according to the provided task instruction. Hence, the pretrained data-driven model may be configured to be provided with a plurality of different task instructions and / or provide a plurality of different types of output data upon receiving different task instructions. The data-driven model may be a finetuned data-driven model, wherein the finetuned data-driven model is obtained by further training a pretrained data-driven model based on historical matching task instructions and corresponding matching scores and / or rankings associated with the historical plant data sets. The finetuned data-driven model(s) may trained additionally on a training data set comprising a plurality of task instructions of one type and corresponding240436

[0048] 8

[0049] output data. The finetuned data-driven model may be trained additionally to provide output data of a predefined type according to the training data set. The finetuned data-driven model may be configured to be provided with a plurality of different task instructions and / or provide a plurality of different types of output data upon receiving different types of task instructions. Further, the finetuned data-driven model may be configured to provide one type of output data upon receiving one type of task instruction with a higher accuracy than providing other types of output data upon receiving other types of task instructions.

[0050] In an embodiment, matching the historical plant data sets with the malfunctional data may comprise determining a matching score per pair of malfunctional data and historical plant data set. The matching score may be related to matching a number of elements associated with the historical plant data sets and the malfunctional data.

[0051] Additionally or alternatively, the matching score may be related to determining a distance between digital representations.

[0052] In an embodiment, any one of the methods may further comprise adapting at least a part of the matching scores associated with the subgroup of historical plant data sets by providing a matching task instruction for determining at least one matching score per historical plant data set The matching task instruction may include the malfunction data, the subgroup of historical plant data sets and the associated matching scores.

[0053] In an embodiment, any one of the methods may further comprise determining one or more cause cluster(s) comprising one or more historical plant data set(s) from the subgroup of historical plant data sets associated with at least one cause related to the plurality of the differences, and wherein providing the one or more cause(s) comprises providing one or more cause(s) per cause cluster. Any one of the methods may further comprise selecting at least one of the cause cluster(s) according to the malfunctional data and providing the cause(s) may comprise providing the at least one cause associated with the selected cause cluster. By doing so, the relevant causes can be provided without any repetition. This improves the human-machine interaction and reduces the amount of data provided to present related causes.

[0054] In an embodiment, any one of the methods may further comprise determining one or more measure cluster(s) per cause cluster comprising one or more historical plant data set(s) from the subgroup of historical plant data sets associated with at least one measure, and wherein providing the one or more cause(s) comprises providing the at least one cause per cause cluster and wherein providing the one or more measure(s) comprises providing the at least one measure per measure cluster. Any one of the methods may further comprise selecting at least one of the measure cluster(s) according to the malfunctional data and providing the measure(s) may comprise providing the at least one measure associated with the selected measure cluster. By doing so, the relevant measures can be provided without any repetition. This improves the human-machine interaction and reduces the amount of data provided to present related measures.In an embodiment, selecting at least a part of the historical plant data sets by matching at least a part of the malfunctional data and the plurality of historical plant data sets may comprise selecting at least the part of the historical plant data associated with the target chemical plant, in particular at least a part of equipment and / or operating instructions associated with the target chemical plant.

[0055] In an embodiment, selecting a subgroup of at least the selected part of the historical plant data sets based on a digital representation of at least the part of the historical plant data sets and a digital representation of the malfunctional data may comprise selecting one or more measure(s) associated with the provided malfunctional data.

[0056] In an embodiment, providing one or more digital representation(s) of at least the selected part of the historical plant data sets and a digital representation of the malfunctional data may be provided via a database, preferably the database comprising the plurality of historical plant data sets. In an embodiment, at least one digital representation of the historical plant data sets may be associated per historical plant data set, in particular at least the selected part of the historical plant data sets. Additionally or alternatively, providing the one or more digital representation(s) of at least the selected part of the historical plant data sets and a digital representation of the malfunctional may comprise determining at least the selected part of the historical plant data sets and a digital representation of the malfunctional by providing at least the selected part of the historical plant data sets and the malfunctional to one or more embedding layer(s). The one or more embedding layer(s) may be configured to generate digital representations of data in response to receiving the data.

[0057] In an embodiment, the digital representations may be provided in response to selecting at least a part of the historical plant data sets and / or together with at least the selected part of the historical plant data sets. The one or more digital representation(s) of at least the selected part of the historical plant data sets may be obtained by providing at least the selected part of the historical plant data sets, ie a predefined number of the historical plant data sets, at once or one data set after another to one or more embedding layer(s). In an embodiment, the one or more digital representation(s) of at least the selected part of the historical plant data sets may be obtained by providing a subsection of at least the selected part of the historical plant data sets to one or more embedding layer(s). The subsection of at least the selected part of the historical plant data sets may be associated with the one or more difference(s) between the historical operation(s) and / or configuration(s) and the target operation and / or configuration. Additionally or alternatively, the subsection of at least the selected part of the historical plant data sets may be associated with the one or more measure(s) and / or cause(s) related to at least the selected part of the historical plant data sets. In an embodiment, the digital representation of the malfunctional data may be associated with and / or may be a representation of the difference between the current operation and / or configuration and the target operation and / or configuration.10

[0058] In an embodiment, selecting at least a part of the historical plant data sets may comprise obtaining, in particular receiving a number of historical plant data sets smaller than a number of available historical plant data sets. Additionally or alternatively, selecting at least a part of the historical plant data sets may comprise providing a ranking of at least the part of the historical plant data sets. The ranking may be indicative of a matching and / or a similarity of at least the part of the historical plant data sets with the malfunctional data. Selecting at least a subgroup of at least the selected part of the historical plant data sets may comprise adapting the ranking associated with a subgroup of at least the selected part of the historical plant data sets.

[0059] In an embodiment, an element may refer to one or more datapoint(s) comprised by the data and / or data set(s). Element may be a part of the historical plant data set(s) and / or the malfunctional data.

[0060] BRIEF DESCRIPTION OF THE SEVERAL VIEWS OF THE DRAWINGS

[0061] In the following, the present disclosure is further described with reference to the enclosed figures. The same reference numbers in the drawings and this disclosure are intended to refer to the same or like elements, components, and / or parts.

[0062] FIG. 1 illustrates an embodiment of an operating system 106 of one or more chemical plants 102.

[0063] FIG. 2 illustrates an embodiment of a method for controlling and / or monitoring an operation and / or a configuration of a target chemical plant.

[0064] FIG. 3 illustrates an embodiment of a method for controlling and / or monitoring an operation and / or a configuration of a target chemical plant.

[0065] FIG. 4 illustrates an embodiment of an operating system 106 of one or more chemical plants 102.

[0066] FIG. 5 illustrates an embodiment of a method for controlling and / or monitoring an operation and / or a configuration of a target chemical plant.

[0067] FIG. 6 illustrates an embodiment of providing one or more cause(s) and / or measure(s) related to malfunctional data for controlling and / or monitoring the one or more target chemical plant(s).11

[0068] FIG. 7 illustrates an embodiment of a user interface for providing malfunctional data and / or providing one or more cause(s) and / or one or more measure(s) for controlling and / or monitoring the target chemical plant(s).

[0069] FIG. 8 illustrates an example of historical plant data sets.

[0070] FIG. 9 illustrates an embodiment of training an embedding layer.

[0071] FIG. 10A illustrates an embodiment of a transformer encoder architecture.

[0072] FIG. 10B illustrates an embodiment of a transformer decoder architecture.

[0073] FIG. 10C illustrates an embodiment of a transformer encoder-decoder architecture.

[0074] FIG. 11 illustrates an embodiment of training and / or deploying the transformer encoder, the transformer decoder and / or the transformer encoder-decoder.

[0075] FIG. 12 illustrates an embodiment of input embedding.

[0076] FIG. 13 illustrates an embodiment of input embedding.

[0077] DETAILED DESCRIPTION

[0078] The following embodiments are mere examples for implementing the method, the system or application device disclosed herein and shall not be considered limiting.

[0079] The following embodiments are mere examples for implementing the method, the system or application device disclosed herein and shall not be considered limiting.

[0080] FIG. 1 illustrates an embodiment of an operating system 106 of one or more chemical plants 102.

[0081] One or more chemical plants 102 may be configured for producing one or more chemical products 104 from raw material 124. The one or more chemical plants 102 may comprise a plurality of components. For example, the components may include heat exchangers, cooling unit, filter, pumps, valves, separation chambers or the like. The one or more components of the one or more chemical plants 102 may be controlled by a control engine. The control engine may be operated by a human worker associated with the chemical plants. Controlling and / or240436

[0082] 12

[0083] monitoring chemical plants may include controlling and / or monitoring a configuration and / or an operation of the chemical plants.

[0084] Configuration of chemical plants may be related to an arrangement of equipment associated with the target chemical plant. For example, the configuration of chemical plants may be related to a type of one or more parts of the equipment and / or a connection between the one or more parts of the equipment. Operation of chemical plants may be related to setting one or more operating parameter(s) associated with at least parts of the equipment for producing chemical products by the chemical plants. Adapting the operating parameters may result in adapting operating of at least a part of the equipment. Configuration and / or operation of chemical plants may be determined from years of experience with operating chemical plants. Nevertheless, incidents associated with chemical plants may occur, e.g. due to human mistakes, chaining of events or unforeseeable circumstances. When an incident occurs, malfunctional data may be generated and / or provided. The malfunctional data may be related to an incident occurring in the target chemical plant. Avoiding any incident allows to save resources while providing higher safety for workers. Thus, it is advantageous to learn from previous incidents.

[0085] To do so, malfunctional data associated with a target chemical plant can be obtained, in particular provided and / or received by an intake interface 112. In an embodiment, the intake interface may comprise a user interface and may be operated by the worker associated with the target chemical plant. In an embodiment, the intake interface may be an interface to a sensor obtaining data related to the operation and / or configuration of the target chemical plant. The malfunctional data may be provided to a measure and / or cause determining engine 110. The measure and / or cause determining engine 110 may be configured to determine one or more measure(s) and / or causes related to the malfunctional data. Hence, the measure and / or cause determining engine 110 may be configured to perform the method described in the context of FIG. 2.

[0086] FIG. 2 illustrates an embodiment of a method for controlling and / or monitoring an operation and / or a configuration of a target chemical plant.

[0087] Configuration of chemical plants may be related to an arrangement of equipment associated with the target chemical plant. For example, the configuration of chemical plants may be related to a type of one or more parts of the equipment and / or a connection between the one or more parts of the equipment. Operation of chemical plants may be related to setting one or more operating parameter(s) associated with at least parts of the equipment for producing chemical products by the chemical plants. Adapting the operating parameters may result in adapting operating of at least a part of the equipment. Configuration and / or operation of chemical plants may be determined from years of experience with operating chemical plants. Nevertheless, incidents associated with chemical plants may occur, e.g. due to human mistakes, chaining of events or unforeseeable circumstances. Avoiding any incident240436

[0088] 13

[0089] allows to save resources while providing higher safety for workers. Thus, it is advantageous to learn from previous incidents.

[0090] Malfunctional data associated with a difference between a current operation of a target chemical plant and a target operation of the target chemical plant and / or a current configuration of the target chemical plant and a target configuration of the target chemical plant may be provided 202. The malfunctional data may be provided via a user interface, e.g. entered by a worker associated with the target chemical plant. This allows to report on incidents and check for measures close in time and directly by workers working in the target chemical plant. Consequently, critical situations may be resolved to avoid further undesired events as a result of a large time gap between the occurrence of the incident and applying countermeasures. Additionally or alternatively, the malfunctional data may be provided upon providing a trigger for providing the malfunctional data. The trigger for providing the malfunctional data may be provided upon detecting a deviation and / or a difference of the operation and / or the configuration of the target chemical plant from a target operation and / or target configuration. The deviation and / or the difference may be detected by a sensor detecting one or more parameter value(s) associated with the operation and / or the configuration of the target chemical plant and comparing the detected one or more parameter value(s) with a predefined parameter value associated with the target configuration and / or operation of the target chemical plant. The predefined parameter value may be specified by a production specification and / or instructions for producing chemical products by the target chemical plant. By doing so, measures to resolve detected operation risks can be taken close in time to avoid critical consequences or worsening of an already critical situation.

[0091] A plurality of historical plant data sets associated with a plurality of differences between historical operations and target operations and / or historical configurations and target configurations of a plurality of chemical plants and a plurality of causes related to the plurality of differences and / or a plurality of measures for eliminating the differences may be provided 204, e.g. by a database. Such a database may be established from historical differences between historical operation and the target operation and / or historical configuration and the target configuration.

[0092] In an embodiment, an indication of a type of the target chemical plant may be provided, e.g. together with the malfunctional data. Based on the indication of the type of the target chemical plant, a group of historical plant data sets associated with the type of the target chemical plant may be selected.

[0093] A matching score per historical plant data set indicative of a matching of the malfunction data and the historical plant data set may be determined 206. The matching score may comprise a numerical value, e.g. between 0 and 1. A higher matching score may indicate a better matching and vice versa. The matching score may be determined based on a distance between numerical representations, in particular vectors, associated with the240436

[0094] 14

[0095] historical plant data sets and a vector associated with the malfunctional data. The distance may comprise an Euclidean distance, a cosine similarity, a Jaccard distance, a minkowski distance or the like. The matching score may be determined with respect to the plurality of historical plant data sets and / or the group of historical plant data sets associated with the type of the target chemical plant. Thereby, a scalable yet accurate matching score can be determined to allow for reliable determination of measures where an indicent in a chemical plant occurred.

[0096] Additionally or alternatively, the matching score may be obtained by carrying out an index search. The malfunctional data and the historical plant data sets may comprise one or more elements. In particular, each historical plant data set may comprise one or more elements. Determining the matching score may include determining the elements associated with the malfunctional data corresponding to the element(s) associated with the plurality of plant data sets. Selecting historical plant data sets may include selecting the historical plant data sets associated with a number of elements corresponding to the one or more element(s) associated with the malfunctional data. Index search is fast and scalable while providing good search results.

[0097] According to the matching score a predefined number of historical plant data sets may be selected 208.

[0098] Additionally or alternatively, the historical plant data sets being associated with a matching score within a predefined matching range may be selected. The predefined matching range may be indicative of one or more numerical values. The predefined number of historical plant data sets may be provided. Additionally or alternatively, the predefined number may be adjusted to

[0099] At least a part of the matching scores associated with the predefined number of historical plant data sets may be adapted by matching a digital representation of the malfunction data with digital representations of the predefined number of historical plant data sets 210. This may include determining second matching scores and exchanging the matching scores determined in 206 by the second matching scores. Matching the digital representation of the malfunctional data with the digital representations of the selected number of historical plant data sets may comprise determining a distance between the digital representations, e.g. as described above. The numerical representations may comprise vectors and / or may be vector representations. This may be known as similarity search. Similarity search compares the content of datasets rather than predefined elements such as words. In particular where the searchable corpus includes string data, matching via context provides the advantage to provide good matching results despite using synonyms. Nevertheless, similarity search requires determining distance per pair of historical plant data sets and malfunctional data and thus, high computational resources. Therefore, a rough filter such as an index search prior to a more sensitive filter towards context allows to save computational resources and thus scale applications while providing accurate results from a similarity search. Hence, by filtering twice, a suited balance between good search and computational resources is enabled.240436

[0100] 15

[0101] A subgroup of historical plant data sets according to the associated matching scores may be selected 212. The subgroup of historical plant data sets may comprise a predefined number of historical plant data sets and / or a number historical plant data sets associated with adapted matching scores within an adapted matching score range. The adapted matching score range may be analogue to the matching score described above.

[0102] At least a part of the matching scores associated with the subgroup of historical plant data sets may be adapted by providing a matching task instruction for determining at least one matching score per historical plant data set, wherein the matching task instruction includes the malfunction data, at least a part of the subgroup of historical plant data sets and the associated matching scores to a data-driven model 214. The data-driven model may be configured to follow task instructions. The matching task instruction may trigger the data-driven model to provide adapted matching scores associated with at least the part of the subgroup of historical plant data sets and the malfunctional data and / or provide a ranking indicative of a matching between at least the part of the subgroup of historical plant data sets and the malfunctional data. The data-driven model may be a data-driven model as described in the context of FIG. 9 to FIG. 13. The data-driven model may comprise one or more embedding layer(s) as described in the context of FIG. 9 or providing the matching task instruction may include mapping the matching task instruction to a digital, in particular numerical, representation of the matching task instruction. The data-driven model may be a language model, in particular a large language model and / or a foundational model. In an embodiment, the data-driven model may be trained based on historical matching task instruction and corresponding rankings of the plant data sets associated with the historical matching task instruction and / or matching scores associated with the plant data sets associated with the historical matching task instruction. Hence, the data-driven model may be configured to match data sets, ie a language model finetuned for the specific purpose to match data sets.

[0103] Optionally, one or more historical plant data sets may be eliminated from the selected subgroup of historical plant data sets upon adapting at least a part of the matching scores by the data-driven model.

[0104] One or more cause cluster(s) comprising one or more historical plant data sets from the subgroup of historical plant data sets may be determined 216. The one or more cause cluster(s) may be associated with at least one cause related to the one or more difference(s) associated with the subgroup of historical plant data sets. Historical plant data sets of one cause cluster may be associated with the at least one cause of the cause cluster. Each historical plant data set of one cause cluster may be associated with the at least one cause. Hence, historical plant data sets associated with one cause cluster may be obtained in relation to one or more difference(s) caused by the cause associated with the cause cluster. Determining the cause cluster(s) may comprise determining matching scores associated with the one or more causes of the subgroup of historical plant data sets. The matching scores may be determined based on vector similarity between pairs of vector representations of at least parts of the historical plant data sets. Hence, a matching score may be determined per pair of the historical plant240436

[0105] 16

[0106] data sets. A cluster may be associated with the at least one cause in response to determining the matching scores associated with the historical plant data sets to be within a predefined cause matching range. For example, a subgroup of 5 historical plant data sets may be selected. Two of the causes associated with the historical plant data set may have a matching score within the predefined cause matching range. Additionally or alternatively, from the remaining 3 historical plant data sets a matching score associated with a first and a second of the remaining 3 historical plant data sets and a matching score associated with the first and a third of the remaining 3 historical plant data sets may be within the predefined cause matching range. Hence, the 5 historical plant data sets may be clustered into two different clusters.

[0107] Additionally or alternatively, determining the cause cluster(s) may comprise providing the cause cluster task instructions for receiving the one or more cause cluster(s) to the data-driven model. The data-driven model may be the same as the data-driven model described in the context of 214 and / or a different data-driven model, in particular associated with a model architecture and / or implementation as described in the context of FIG. 9 to FIG.

[0108] 13. The cause cluster task instructions may comprise at least a part, in particular a predefined number, of the subgroup of historical plant data sets. Additionally or alternatively, an index and / or a similarity search may be performed in relation to at least the part of the subgroup of historical plant data sets.

[0109] One or more measure cluster(s) per cause cluster comprising one or more data sets from at least a part of the subgroup of historical plant data sets may be determined 218. The measure cluster(s) may be associated with at least one of the plurality of measures associated with at least the part of the historical plant data sets. The measure cluster(s) may be obtained analogous to the cause cluster(s). At least one measure cluster task instruction may be provided per cause cluster. The measure cluster task instruction for clustering one or more historical plant data sets associated with one cluster may comprise the one or more historical plant data sets associated with one cluster.

[0110] Optionally, selecting at least one cause cluster according to the one or more matching score(s) associated with at least the part of the subgroup of historical plant data sets related to one or more cause cluster(s). For example, the cause cluster associated with the highest matching score may be selected.

[0111] The one or more measure(s) in relation to the one or more causes may be provided 220. This may include providing the one or more cause cluster(s) and / or the one or more measure cluster(s). Preferably, the one or more cause cluster(s) and the one or more measure cluster(s) per cause cluster may be provided. For example, a list of potential causes in relation to the malfunctional data and a list of potential measures per cause may be provided. The one or more measure(s) and / or the one or more cause(s) in relation to the determined cluster(s) may be provided, e.g. via an user interface. The one or more cause(s) and / or measure(s) may be provided via a user interface, in particular displayed to, a worker associated with the target chemical plant. This allows to implement240436

[0112] 17

[0113] measures and / or resolve causes related to operation and / or configuration of chemical plant controlled by a human worker.

[0114] FIG. 3 illustrates an embodiment of a method for controlling and / or monitoring an operation and / or a configuration of a target chemical plant.

[0115] The method for controlling and / or monitoring the operation and / or configuration of the target chemical plant may comprise a preselection of the historical plant data sets according to the malfunctional data via an index and / or keyword search, compare 206 and / or 208. By doing so, a matching score per historical plant data set may be determined. The historical plant data sets may be ranked according to the matching score associated with the historical plant data sets. A more accurate searching strategy may be applied to adapt the ranking obtained by the index search via a vector and / or embedding search, compare 210 and / or 212. Vector and / or embedding search may require higher computational cost. Hence, this type of search can be scaled to lower dimensions than the index search. As a consequence, vector and / or embedding search may be applied to a part of the historical plant data sets including less historical plant data sets than all provided historical plant data sets. This may allow to apply the more accurate search strategy to the most relevant documents. This relevant subsection of documents may be more similar to each other than less relevant documents may be similar to the more relevant (and thus selected) part of historical plant data sets. Followingly, accurate recognition of topics associated with the historical plant data sets may be conducted more accurately while increasing time of retrieval. As described in the context of FIG. 1 and 2 this may ultimately increase the efficiency of controlling and / or monitoring chemical plants.

[0116] In an embodiment, two or more historical plant data sets may be associated with the same cause and / or measure. Hence, it may be desired to provide one cause and / or measure once to reduce unnecessary data transfer. This can be achieved by clustering the subgroup of historical plant data sets according to the causes and / or measures associated with the subgroup of historical plant data sets. Further details on the clustering are described in the context of 216 and / or 218. At least one cause and / or measure per cluster can be provided e.g. via a user interface to allow for monitoring and / or controlling of the target chemical plant.

[0117] FIG. 4 illustrates an embodiment of an operating system 106 of one or more chemical plants 102.

[0118] The operating system 106 of the one or more chemical plants 102 may be as described in the context of FIG. 1. The measure and / or cause determining engine 110 may comprise an index search engine 402, an embedding search engine 404, a cause clustering engine 406 and / or a measure clustering engine 408.

[0119] The index search engine 402 may be configured to determine a matching score per historical plant data set indicative of a matching of the malfunction data and the historical plant data set and / or select a predefined number240436

[0120] 18

[0121] of historical plant data sets according to the associated matching scores. The index search engine 402 may be configured to carry out the steps 206 and / or 208 as described in the context of FIG. 2.

[0122] The embedding search engine 404 may be configured to adapt at least a part of the matching scores associated with the predefined number of historical plant data sets and malfunctional data by determining matching scores associated with the predefined number of historical plant data sets and malfunctional data and / or select a subgroup of historical plant data sets according to the associated matching scores. The embedding search engine 404 may be configured to carry out the steps 210 and / or 212 as described in the context of FIG. 2.

[0123] The cause clustering engine 406 may be configured to determine one or more cause cluster(s) comprising one or more data sets from the subgroup of historical plant data sets. Hence, the cause clustering engine 406 may be configured to cluster the subgroup of historical plant data sets according to the causes associated with the subgroup of historical plant data sets. The cause clustering engine 406 may be configured to carry out the step 214 as described in the context of FIG. 2.

[0124] The measure clustering engine 408 may be configured to one or more measure cluster(s) per cause cluster comprising one or more data sets from the subgroup of historical plant data sets according to the measure(s) associated with the historical plant data sets. Hence, the measure clustering engine 408 may be configured to cluster the historical plant data sets per cluster according to the measures associated with the subgroup of historical plant data sets. The measure clustering engine 408 may be configured to carry out the step 216 as described in the context of FIG. 2.

[0125] Optionally, the measure and / or cause determining engine 110 may further comprise a model engine (not shown). The model engine may be configured to adapt the matching score determined by the embedding search engine 404 and / or index search engine 402 as described in the context of FIG. 2.

[0126] FIG. 5 illustrates an embodiment of a method for controlling and / or monitoring an operation and / or a configuration of a target chemical plant.

[0127] The malfunctional data may be provided by a user, in particular a worker associated with the one or more target chemical plant(s) as described earlier. A request for receiving a selection of historical plant data sets may be provided by an application to a database comprising a plurality of historical plant data sets. The selection of the historical plant data sets may comprise a ranking of the historical plant data sets according to matching scores indicative of a similarity between the historical plant data sets and the malfunctional data. The database may be configured to determine a predefined number of historical plant data sets corresponding at least in parts to the malfunctional data. The historical plant data sets may be associated with a digital representation of the historical240436

[0128] 19

[0129] plant data sets. These digital representations may be optionally provided by the database. Alternatively, the digital representations of the historical plant data sets may be generated by one or more embedding layer(s). The embedding layer(s) may be configured to generate a digital representation of data from data provided to the embedding layer(s). The embedding layer(s) may be as described in the context of FIG. 9.

[0130] From the predefined number of historical plant data sets adapted matching scores may be determined based on the digital representations of the historical plant data sets, e.g. as described in further detail in the context of FIG.

[0131] 2. Based on the adapted matching scores, a subgroup of historical plant data sets may be selected. The subgroup may be clustered according to the causes associated with the historical plant data sets. Further, per cause cluster the historical plant data sets may be clustered according to the measures associated with the historical plant data sets. Thereby, at least one measure and / or cluster may be determined per cause cluster and / or measure cluster. The determined causes and / or measures per cluster may be determined to the user, e.g. via a user interface as described in the context of FIG. 7.

[0132] FIG. 7 illustrates an embodiment of a user interface for providing malfunctional data and / or providing one or more cause(s) and / or one or more measure(s) for controlling and / or monitoring the target chemical plant(s).

[0133] The user interface input 702 may allow to enter a plant ID. Hence, the malfunctional data may include a digital indentifier associated with the one or more target chemical plant(s). Further, the user interface may allow to enter the observed result of the incident. In some cases, a cause, e.g. being a result of the root cause, or a misfunctioning of the production of the one or more target chemical plant(s) may be detected. This may be reported as the result of the incident. The root cause may be typically not observed directly and thus, evaluated by the measure and / or cause determining engine 110, e.g. operated via the user interface input 702. To allow for better evaluation of the root cause and related measures, sensor data collected by monitoring sensors associated with at least a part of the one or more target chemical plant(s), in particular associated with equipment of the one or more target chemical plant(s) related to the incident occurred may be provided.

[0134] Preferably, where the malfunctional data may be provided by a human user, the malfunctional data may comprise string data, in particular one or more word(s) preferably forming one or more sentence(s). Thereby, the measure and / or cause determining engine 110 can be operated by worker associated with the one or more target chemical plant(s). Hence, implementation of measures and / or elimination of root causes is accelerated.

[0135] FIG. 7 illustrates an embodiment of a user interface for providing malfunctional data and / or providing one or more cause(s) and / or one or more measure(s) for controlling and / or monitoring the target chemical plant(s).240436

[0136] 20

[0137] By the measure and / or cause determining engine 110 as described above, related incident reports with details on causes and measures as well as a summary of potential root causes and related measures may be provided via the user interface input 702, e.g. to a worker associated with the one or more target chemical plant(s). Preferably, the provided root cause and measures may include human-interpretable data, in particular text data and / or image data to allow for the workers to resolve the root cause and apply the measure effectively.

[0138] FIG. 8 illustrates an example of historical plant data sets.

[0139] FIG. 9 illustrates an embodiment of obtaining an embedding layer. The embedding layer may be obtained by training for example a continuous bag of words model (CBOW) or a skip-gram model. The embedding layer may be suitable for generating embedded input data based on input data. Generating embedded input data may refer to embedding input data. Embedding input data may result in a representation associated with the input data. Thus, the embedded input 914 may be the representation associated with the input data. The input data may comprise one or more elements. The one or more elements may be represented by the input vector 906. In particular, the embedded input 914 and / or the input vector 906 may be machine- readable and / or processable by a processor. For this purpose, the embedded input 914 and / or the input vector 906 may be a tensor, in particular a first-rank tensor. Specifically, the input vector 906 may be a one-hot vector or a summation of a plurality of one-hot vectors. A one-hot vector may be a vector with one entry unequal to zero. Examples for one-hot vectors may be 908, 910 and 912. The entries unequal to zero in the one-hot vector and / or in the input vector 906 may indicate the element. For example, a lookup table may define the relation between the position of the entries unequal to zero and the element indicated by the one-hot vector. The lookup table may specify a plurality of different elements. The number of different elements may be equal to the number of entries in the one-hot vector. The number of different elements may be referred to as vocabulary size. In an example, the elements may be represented by tokens and a sequence of elements may refer to at least a part of a sentence. The at least a part of the sentence may be represented by a plurality of tokens. A token may represent at least a part of the element and / or word. For example, where one element would be associated with only one word, words such as “embeddings", “embedding” or “embed” would constitute different elements. A first token may represent the stem “embed” and the endings, typically appearing in a plurality of word, may be represented by a second token, a third token and a fourth token. The second token, the third token and the fourth token may be used for representing other words such as “look”, “looking” or the like, preferably together with a fifth token representing the stem “look”. Ultimately, this tokenization of elements associated with a plurality of stems and a plurality of endings results in less tokens to be used for representing a plurality of elements and thus, uses less computational resources. A lookup table specifying a subset of the vocabulary size e.g. of the English language may comprise 10,000 words or more. The embedded input 914 may be a lower-dimensional representation than the input vector 906. For example, typical embedded inputs 914 may comprise some hundreds of different entries. Followingly, the embedded inputs 914 constitute a densified representation of one or more elements using less computationalresources. More than that, the embedded input 914 may represent a relation between two or more elements. For example, the words “Italy” and “Germany” may be similar or may be more closely related since they both define European countries, whereas the word “embodiment” may be very different from the two respective words. The smaller the dot product between two embedded inputs 914 may be the more similar the two elements associated with the embedded inputs 914 may be. Hence, the embedded inputs 914 may represent one or more elements accurately and lead to accurate results based on processing the embedded inputs 914.

[0140] For transforming the input vector 906 into the embedded input 914, the embedding layer may comprise a number of neurons equal to the number of entries in the embedded input 914. Based on the embedded inputs 914, the output layer may generate the output vector 916. The output vector may be a vector and / or may indicate one or more elements. The output vector 916 may indicate one or more elements different from the input vector 906 and / or the one-hot vectors associated with the input vector 906. For this purpose, the output layer may comprise a number of neurons equal to the number of entries of the input vector 906 and / or the output vector 916. The output layer may apply a softmax function to the embedded inputs 914. By doing so, the output vector may comprise the probabilities associated with the elements associated with the entries of the output vector 916 unequal to zero. Hence, from the output vector 916 one or more elements may be obtained with a corresponding probability. Where the input vector 906 may specify one or more sequence(s) of elements, the output vector 916 may specify one or more elements corresponding to the sequence(s) of elements specified by the input vector 906. In the example of FIG. 9, the element associated with vector 918 may correspond to the input vector with a probability of 71 %. Additional or alternative elements may correspond to the input vector as indicated by the output vector with lower probability. By defining a threshold to which the probability may be compared, the selection of the corresponding elements may be tailored to the needs of the user. The elements generated by the model comprising the embedding layer 902 and the output layer 904 may refer to the most probable elements indicated by the output vector 916. Hence, the model depicted in FIG. 9 may generate the element associated with the vector 918 with a confidence score of 71 %.

[0141] The model of FIG. 9 may be continuous bag of words (CBOW) model. The CBOW model may be trained based on a training data set comprising a plurality of input vectors and corresponding output vectors. As the training data set may not be labeled, the training of the CBOW model may be referred to as self-supervised. Before training of the CBOW model, the CBOW model may be initialized with random values assigned to the weights of the neurons. During the training of the CBOW model, the input vectors may be passed through the initialized embedding layer and the output layer and a loss may be determined by comparing the output vector obtained by passing the input vector 906 through the model to the output vector corresponding to the input vector 906 as specified by the training data set. Based on the determined loss, backpropagation may be applied to determine the gradients associated with the neurons of the embedding layer 902 and the output layer 904 to lower the loss. According to the determined gradients, the weights of the neurons may be updated by using a gradient descent algorithm. If a predetermined loss may be achieved by the CBOW model, the training may be terminated and a trained CBOW model may be obtained. From the trained CBOW model, the embedding layer 902 may be suitable for embedding22

[0142] input data comprising one or more elements. This embedding layer 902 may be used in other machine-learning architectures requiring an embedding layer 902 such as a transformer encoder, transformer decoder or transformer encoder decoder architecture as described within the context of FIG. 10A, FIG. 10B and FIG. 10C. For training these architectures, a trained embedding layer 902 may be required. Hence, a model such as a CBOW model may be trained prior to training the transformer encoder, transformer decoder or transformer encoder decoder architecture.

[0143] FIG. 10A illustrates an embodiment of a transformer encoder architecture. The transformer encoder comprises an encoder input 1078, one or more encoder blocks 1074, 1014 and an encoder output. The transformer encoder architecture may be derived from the transformer encoder-decoder architecture as known in the art and shown in FIG. 10C. In particular, the transformer encoder may be referred to as X-former. The transformer encoder architecture may correspond to the encoder architecture associated with the transformer encoder-decoder architecture with an additional encoder output instead of connecting the encoder block directly to the decoder of the transformer encoder-decoder architecture. A plurality of transformer encoder architectures are available in the art such as the bi-directional encoder representations from transformers (BERT).

[0144] The input data may be received at the encoder input 1078. The encoder input 1078 may apply an input embedding 1002. Applying the input embedding 1002 may refer to passing the input data through an embedding layer eg as described within the context of FIG. 9. Further, the encoder input 1078 may apply positional encoding 1004. Applying positional encoding 1004 may refer to adding a positional factor to the embedded input obtained via input embedding. Preferably, the input data may specify a sequence of elements. The positional factor Pp°smay be indicative of the position of the elements within the sequence. For example, the positional factor Pp°smay be obtained based on the following equation:

[0145]

[0146] where pos may refer to the position of the element within the sequence, / may refer to the dimension associated with the input embedding and d may refer to the dimension of the model, e.g. transformer decoder, transformer encoder or transformer encoder-decoder. This may be referred to as absolute positional embeddings.

[0147] Alternatively, the positional encoding may be based on rotary positional embeddings (RoPE). Positional encoding is beneficial since it enables the processing of sequential data without requiring further dimensions indicating the position of each element. Followingly, the positional encoding 1004 reduces the computational resources needed for embedding the input data. By passing the input data through the encoder input, the input data may be transformed into a second-rank tensor representing the sequence of elements. This second-rank tensor may be23

[0148] referred to as embedded input data. The embedded input data may be processed by the encoder block. The embedded input data may be provided to the layer normalization 1008 by a residual connection. Multi-head selfattention 1006 may be applied to the embedded input data. Multi-head self-attention 1006 may comprise the two components multi-head and self-attention. Self-attention may be understood as being a filter applied to the embedded input data. By applying the filter to the embedded input data, the elements associated with the embedded input data contributing to the to be generated output data may be identified for generating the output data. Hence, the filter may represent the degree of contributing to the to be generated output data by the elements associated with the embedded input data. Applying the filter may be referred to as weighting the elements associated with the embedded input data. This is advantageous specifically regarding long sequences of elements. The filter may be learned and improved during the training by learning to identify the contribution of elements associated with the embedded input data. For example, in the partial sentence “I went to the bakery to buy a” the last word may be generated by the data-driven model such as the transformer encoder. The selfattention may focus the transformer encoder to attend to the word “bakery” and “buy” mostly to generate the word “bread”. Self-attention may refer to attention generated based on the input data. Hence, the filter may be determined based on the input data, preferably the embedded input data. The embedded input data may serve as query Q, key K and value V with respect to the self-attention operation. The self-attention may refer to attention based on the received input data. Hence, the filter may be calculated based on the following formula by inserting the respective tensors based on the embedded input data:

[0149]

[0150] where dk corresponds to the dimension of the key.

[0151] For improving the efficiency of the transformer encoder further, the multiple heads are used to apply the filter resulting in the multi-head self-attention 1006. Multi-head self-attention 1006 may comprise applying the filter to two or more parts of the embedded input data. Hence, the tensor may be split into two or more parts and the filter may be applied to the two or more parts separately by two or more heads according to the

[0152] following equation: head i = Attention (QW,Q, KW<K, VW,v'}

[0153] with parameter matrices jy.Q jjdxdrwhere I may refer to the number

[0154]

[0155] of heads,

[0156]

[0157] may refer to the dimensions of the value, key and query.

[0158] The result of the two or more head may be concatenated according to the following

[0159] equation:

[0160]

[0161] e-^hdv*d and h may refer to the number of heads.

[0162]

[0163] The embedded input data may be transformed via the multi-head self-attention 1006 into a context tensor. The context tensor may represent the sequence of elements and the relation between two or more elements of the24

[0164] input data. The context tensor may be a second rank tensor and / or may comprise one or more first rank tensor(s). After the multi-head self-attention 1006 layer normalization 1008 may be applied based on the context tensor and / or the embedded input data from the residual connection. Applying layer normalization 1008 may refer to normalizing the context tensor. Normalizing the context tensor may lower the values of the entries of the context tensor. This reduces the computational cost associated with processing the context tensor. Further, it improves the training by contributing the loss to converge and preventing instabilities.

[0165] Layer normalization 1008 may be followed by passing the context tensor to a feed-forward layer 1010 again followed by layer normalization 1012 based on the residual connection to the context tensor and / or the output of the feed-forward layer 1010. The feed-forward layer 1010 may be a feed-forward neural network. The feedforward neural network may comprise of a plurality of fully connected neurons. Passing the context tensor through the feed-forward neural network may result in transforming the context tensor linearly. Additionally or alternatively, the neural network may comprise one or more activation functions such as a rectified linear unit (ReLU). Hence, the neural network may be configured for performing one or more non-linear operations to the context tensor and / or transforming the context tensor non-l inearly . After the context tensor has been transformed and / or normalized by the feed-forward layer 1010 and the layer normalization 1012, the context tensor may be provided to one or more further encoder blocks 1014. Having passed the context tensor through the feed-forward layer 1010 may adapt the context tensor for the processing by a further attention layer of the one or more further encoder blocks 1014 for applying a self-attention filter, preferably multi-head self-attention 1006. The context vector after being transformed by the layer normalization 1012 and the feed-forward layer 1010 may be referred to as hidden state.

[0166] The encoder output 1076 comprises of a linear layer 1016 and a softmax layer 1018. The linear layer 1016 may transform the context vector into a logits vector. The linear layer may be fully-connected. The logits vector obtained by passing the context tensor through the linear layer 1016 may be passed through the softmax layer 1018. Passing the logits vector through the softmax layer 1018 may refer to applying the softmax function to the logits vector. Applying the softmax function to the logits vector may result in a probability distribution of one or more elements corresponding to the sequence of elements in the input data. From the probability distribution based on predefined selection criteria, one or more elements may be chosen. The one or more chosen elements may be referred to as the one or more elements generated by the transformer encoder. The one or more generated elements may be provided to the encoder input for generating further one or more elements corresponding to the sequence of the input data and the one or more elements generated by the transformer encoder as described within the context of FIG. 11.

[0167] FIG. 10B illustrates an embodiment of a transformer decoder architecture.

[0168] The transformer decoder comprises a decoder input 1084, one or more decoder blocks 1080, 1032 and a decoder output 1092. The transformer decoder architecture may be derived from the transformer encoder-decoder architecture as known in the art and shown in FIG. 10C. The transformer decoder may be referred to as X-former. The transformer decoder architecture may correspond to the decoder architecture associated with the transformerencoder-decoder architecture independent of receiving one or more hidden states from the encoder of the transformer encoder-decoder. A plurality of transformer decoder architectures are available in the art such as the generative pretrained transformers (GPT).

[0169] The decoder input 1084 may apply input embedding 1020 and positional encoding 1022 analogous to analogous to the input embedding 1002 and the positional encoding 1004 as described within the context of FIG. 10A. The decoder block 1080 may comprise the layer normalizations 1026, the masked multi-head self-attention 1024, the feed-forward layers 1028 and / or the layer normalization 1030. The embedded input data resulting from passing the input data through the decoder input 1084 may be provided to the layer normalization 1026 via a residual connection. Further, masked multi-head self-attention 1024 may be applied to the embedded input data. Masked multi-head self-attention 1024 corresponds to the multi-head self-attention 1006 as described within the context of FIG. 10A with additionally masking a part of the embedded input data associated with elements later in the sequence than the element to be generated. Additionally or alternatively, the part of the input data associated with elements later in the sequence than the element to be generated may not be received and / or transformed into the embedded input data. Thus, the transformer decoder may be suitable for generating a subsequent element to a sequence, whereas the transformer encoder may be suitable for generating a missing element in within one sequence and / or between two or more sequences. Therefore, the transformer encoder may be configured for classification tasks. The transformer decoder may be configured for text generation.

[0170] Similar to the transformer encoder as described within the context of FIG. 10A, a context tensor may be generated by applying the masked multi-head self-attention 1024 and the layer normalization 1026. The context tensor may be provided to the layer normalization 1030 via a residual connection. Further, the feed-forward layer 1028 and the layer normalization 1030 may be analogous to the feed-forward layer 1010 and the layer normalization 1012 as described within the context of FIG. 10A. The context tensor may be provided to one or more further decoder blocks 1032.

[0171] The decoder output 1092 may comprise of a linear layer 1034 and a softmax layer 1036. The linear layer 1034 and the softmax layer 1036 may be analogous to the linear layer 1016 and the softmax layer 1018 as described within the context of FIG. 10A.

[0172] FIG. 10C illustrates an embodiment of a transformer encoder-decoder architecture. The transformer encoderdecoder may comprise the encoder input 1088, the one or more encoder blocks 1086, 1064, the decoder input 1094, the decoder block 1090 and the decoder output 1092. The encoder input 1088 may correspond to the encoder input 1078 of FIG. 10A. The one or more encoder block 1086, 1064 may correspond to the one or more encoder blocks 1074, 1014 of FIG. 10A. The decoder input 1094 may correspond to the decoder input 1084 of FIG. 10B.

[0173] The decoder block 1090 may comprise a masked multi-head self-attention 1070, a layer normalization 1072, a feed-forward layer 1038 and a layer normalization 1040 analogous to the masked multi-head self-attention 1024, the layer normalization 1026, the feed-forward layer 1028 and the layer normalization 1030 as described within the context of FIG. 10B. The decoder block 1090 may further comprise a multi-head self-attention 1050 and a26

[0174] layer normalization 1048. Analogous to the description of FIG. 10B, the context tensor may be obtained from the masked multi-head self-attention 1070 and the layer normalization 1072. Multi-head self-attention 1050 analogous to the multi-head self-attention 1006 of FIG. 10A may be applied to the context vector obtained from the layer normalization 1072 and the hidden states of the one or more encoder blocks 1086, 1064. Layer normalization 1048 may be applied to the context vector obtained from the multi-head self-attention 1050 and the context vector obtained from the layer normalization 1072 provided via a residual connection. The context vector resulting from the layer normalization 1048 may be processed via the feed-forward layer 1038 and the layer normalization 1040 analogous to the description of FIG. 10B. The context vector resulting from the layer normalization 1040 may be provided to further decoder blocks 1042 analogous to the decoder block 1090. The context vector obtained from the one or more decoder blocks 1090, 1042 may be provided to the decoder output 1092. The decoder output 1092 may correspond to the decoder output 1082 of FIG. 10B.

[0175] With the above-described architecture, the transformer encoder-decoder may receive and process input data at the encoder input 1088 and the one or more encoder blocks 1086, 1064 and the decoder block 1090 and the decoder output 1092. Based on the input data, the transformer encoder-decoder may generate output data part by part or sequentially. The sequentially generated output data may be provided to and / or may be processed by the decoder input 1094, the one or more decoder blocks 1090, 1042 and the decoder output 1092. Preferably, a sequence may be provided to the encoder input 1088 and after having generated at least a part of the output data, the decoder input 1094 may be provided with at least the part of the elements of the output data already generated. By doing so, the next elements of the output data may be generated with a higher accuracy by taking the input data and the generated output data into account since more data is received by the transformer encoderdecoder may be received over time.

[0176] Because of the transformer encoder-decoder architecture, the transformer encoder-decoder may be configured for transforming a sequence into another representation of the sequence. An example for transforming one sequence into another representation may be translation of one sentence into another language. A plurality of transformer encoder-decoders are available in the art such as BART, T5 or the like.

[0177] In an embodiment, the layer normalization 1008, 1012 may be applied prior to the masked multi-head selfattention 1024, multi-head self-attention 1006 and / or the feed-forward layer 1010 in the transformer decoder, the transformer encoder and / or the transformer encoder-decoder. By doing so, the computational resources for applying the multi-head self-attention 1006 and / or the feed-forward layer 1010 to the embedded input data and / or the context tensor may be decreased as the entries of the respective tensors may be lower after normalization. In an embodiment, the decoder output 1092 may comprise of a classification neural network, further feedforward layers, convolutional layers, fully connected layers or the like. For example, the transformer encoder-decoder may be configured for choosing between a plurality of options. For this purpose, the transformer encoder-decoder may be provided with three different input data sets and may classify the context vectors obtained from the one or more decoder blocks 1090 via one or more linear layers. Followingly, the architecture may be extended depending on the use case to be solved. [1]27

[0178] FIG. 11 illustrates an embodiment of training and / or deploying the transformer encoder, the transformer decoder and / or the transformer encoder-decoder.

[0179] The encoder / decoder / encoder-decoder architecture 1102 may correspond to the transformer decoder, the transformer encoder and / or the transformer encoder-decoder as describe within the context of FIG. 10A- FIG. 10C.

[0180] The output data generated by the encoder / decoder / encoder-decoder architecture 1102 may comprise of one or more elements, in particular a sequence of elements. The previously generated elements of the output data may be provided as input for generating the next element in the sequence of the output data.

[0181] In the example of FIG. 11, the input data may comprise of N elements, in particular input tokens. An input token may be a token dedicated to be inputted into a data-driven model such as the transformer decoder, the transformer encoder or the transformer encoder-decoder. The output data to be generated may comprise of M elements. The encoder / decoder / encoder-decoder architecture 1102 may generate one element of the output data based on receiving the input data and optionally previously generated elements of the output data at a timestep. Hence, for generating M elements M time steps are required. A time step comprises of providing input 1110, 1112, 1114 to the encoder / decoder / encoder-decoder architecture 1102 and receiving output data 1104, 1108, 1106 from the encoder / decoder / encoder-decoder architecture 1102. In a first timestep, the input 1110 may comprise of N input tokens. The N input tokens may be associated e.g. with N words, stems or endings. Preferably, the N input tokens may specify a question. One or more input tokens may specify the beginning of the sequence of tokens and / or the end of the sequence of tokens. The input 1110 may be processed by the encoder / decoder / encoder-decoder architecture 1102. Based on the input 1110 at least a part of the output data 1104 may be generated. The at least a part of the output data may comprise a first output token. In the next timestep, the generated first output token may be provided together with the input 1112. Specifically, where the input 1112 may be received by a transformer encoder-decoder the input tokens may be received at the encoder input 1088 and the first output token may be received at the decoder input 1094. Where the input 1112 may be received by the transformer encoder, the input 1112 may be received by the encoder input 1078 and analogously regarding the transformer decoder and the decoder input 1084. Based on the input 1112, the output data 1108 comprising the first output token and a second output token may be generated. Generating the output data 1108 based on the input 1112 may refer to generating the second token based on the first token and the N input tokens, wherein the first token may have been generated based on the N input tokens. This process may be repeated until the last token in the sequence of the output data 1106 may be generated. Preferably, the last token may be an end token. The end token may terminate the generation of a further output token.

[0182] Similarly, to the data processing during deployment of the encoder / decoder / encoder-decoder architecture 1102, the encoder / decoder / encoder-decoder architecture 1102 may be trained. The training data set may comprise a plurality of sequences comprising a plurality of elements. The sequences may be associated with the input data and / or the output data. Additionally or alternatively, the sequences may be independent of the input data and / or the output data. For example, where the input data and the output data may refer to chemical compositionsrepresented via text, the training data set may comprise sequential text data independent of chemical compositions. In this example, the training data set may comprise sequences of words originating from a conversation. In an embodiment, the training data set may comprise at least partially input data sets and / or output data sets.

[0183] The training may be initialized by initializing the encoder / decoder / encoder-decoder architecture 1102. In an embodiment, the parameters associated with the encoder / decoder / encoder-decoder architecture 1102 may be initialized randomly. Additionally or alternatively, the input embedding of the encoder / decoder / encoder-decoder architecture 1102 may be obtained by training a CBOW model or a skip gram model as described within the context of FIG. 9. The trained embedding layer may be used during training. The parameters associated with the embedding layer may be kept constant and / or may be updated after a predefined number of training epochs. By doing so, the number of parameters to be updated is lower enabling a faster and less computational resources-consuming training. Further, the accuracy associated with the embedding layer may be constant and / or may be increased by avoiding error compensation in relation to the just initialized encoder / decoder / encoder-decoder architecture 1102.

[0184] During the training of the encoder / decoder / encoder-decoder architecture 1102, at least a part of the sequences of the training data set may be provided to the encoder / decoder / encoder-decoder architecture 1102 one by another and one or more elements may be generated based on the sequences of the training data set one by another. The elements generated based on the sequences may follow the elements of the parts of sequences the encoder / decoder / encoder-decoder architecture 1102 may have been provided with. The generated one or more elements may be compared to the one or more elements following the at least a part of the sequences provided to the encoder / decoder / encoder-decoder architecture 1102 as specified by the training data set. Hence, during the training the encoder / decoder / encoder-decoder architecture 1102 may generate a guess on the next element and the guess on the next element in a sequence may be compared to the ground truth specifying the actual next element according to the training data set. Based on the guess on the next element and the ground truth a loss may be determined. The loss may define the similarity between the guess on the next element and the ground truth. The loss may be determined by forming a vector dot product between the token associated with the one or more elements and the token associated with the ground truth. A loss unequal to zero may result in updating the parameters associated with encoder / decoder / encoder-decoder architecture 1102. Preferably the parameters associated with the encoder / decoder / encoder-decoder architecture 1102 may be independent of the embedding layer. For example, the parameters associated with the encoder / decoder / encoder-decoder architecture 1102 may be weights of the neurons of the encoder / decoder / encoder-decoder architecture 1102.

[0185] Based on the determined loss, backpropagation may be applied to determine the gradients associated with the parameters of the parameters associated with encoder / decoder / encoder-decoder architecture 1102 to lower the loss. According to the determined gradients, the parameters associated with the encoder / decoder / encoder-decoder architecture 1102, preferably the weights of the neurons associated with the encoder / decoder / encoder-decoder architecture 1102, may be updated by using a gradient descent algorithm.29

[0186] The training data set may be unlabeled. The sequences of elements within the training data set may inherently comprise the ground truth for determining the loss with respect to the one or more elements generated during the training of the encoder / decoder / encoder-decoder architecture 1102. Hence, the encoder / decoder / encoder-decoder architecture 1102 may be trained self-supervised. This is advantageous since time and resources for creating a labeled training data set may be saved. Furthermore, this enables the usage of large training data sets associated with a size of several tera bytes. Consequently, the data-driven model may be accurate in generating elements of a sequence. In addition, the large training data set enables few shot predictions or even zero shot predictions. Hence, the data-driven models trained as described above are versatile contributing to saving resources needed for training and / or hosting a plurality of purpose-driven models such as convolutional neural networks. The training described above may be referred to as pretraining. The data-driven model may be configured for performing few shot or even zero shot predictions with respect to a plurality of use cases after pretraining. The performance of the data-driven model may be increased further by additional training referred to as finetuning.

[0187] FIG. 12 illustrates an embodiment of input embedding. Where the sequence of elements associated with the input data, preferably comprised in the input data, may be of one type, the input embedding 1002, 1020, 1052, 1066 as described within the context of FIG. 10A - 2C may be used. For example, a type of input data may be text where the elements may be associated with at least a part of a word, a punctuation character, a start token specifying the beginning of one or more sequences associated with the input data and / or the end token. In another example, the input data may be at least partially numerical. Hence, the input data may comprise a plurality of numbers. Numerical input data may be for example tabular data. Tabular data may specify one or more rows and / or one or more columns. Hence, the tabular data may comprise one or more cells, wherein the cells may be associated with one or more numerical values.

[0188] Numerical input data may require a different embedding than text input data. Input embeddings for numerical input data may comprise a token embedding, a positional embedding, a column embedding, a row embedding or a combination thereof.

[0189] Applying a token embedding to one or more elements, in particular tokens associated with the input data may result in a machine-processable representation associated with the one or more elements, in particular tokens. Applying the token embedding to one or more elements may refer to passing the one or more elements through the embedding layer, e.g. as described within the context of FIG. 9. Hence, token embeddings may specify the one or more elements, in particular tokens in a machine-processable representation. For example, the token embedding may transform a numerical value into a vector. This is advantageous since this representation can be enriched by further information such as the position of the token within the sequence and / or within a table associated with the sequence of tokens. The positional embedding may be analogous to the positional embedding as described within the context of FIG. 9, FIG. 10A-2C. Where the input data may be tabular data, column embedding may be applied. Applying a column embedding to one or more elements, in particular tokens associated with the input data may result in a machine-processable representation specifying the location of the30

[0190] one or more elements within a table 1202, preferably within the columns of the table 1202. Applying the column embedding may refer to adding a column factor to the input data embedded via token embeddings, in particular the embedded input data. The column factor may be the same for elements associated with the same column and / or may differ between two or more elements associated with different columns. Analogous, row embeddings may be applied where the input data may be tabular data. Applying a row embedding to one or more elements, in particular tokens associated with the input data may result in a machine-processable representation specifying the location of the one or more elements within a table 1202, preferably within the rows of the table 1202. Applying the row embedding may refer to adding a column factor to the input data embedded via token embeddings, in particular the embedded input data. The row factor may be the same for elements associated with the same row and / or may differ between two or more elements associated with different rows.

[0191] In an embodiment, input data may be at least partially numerical and at least partially text. Hence, the input data may comprise two or more types of data. A type of data may refer to a modality. Followingly, different embeddings may be applied to the input data. To parts of the input data comprising text the input embedding referred to in FIG.

[0192] 9, FIG. 10A-2C may be applied. To parts of the input data being numerical token embeddings, positional embeddings, column embeddings and row embeddings may be applied. Further, segment embeddings may be applied to the input data independent of the type of input data. The segment embedding may specify the type of input data one or more elements may be associated to. For example, if the input data comprises of text and numbers, the input data may comprise of two types of input data. Applying the segment embedding to the input data may refer to adding a segment factor to the input data, preferably the embedded input data and / or the input data after having applied the token embedding. The segment factor may specify the type of data associated with the one or more elements. The segment factor may be the same for one or more elements associated with the same type of input data and / or may differ between two or more elements associated with different types of input data.

[0193] Applying the token embedding, the positional embedding, the segment embedding, the column embedding, the row embedding or a combination thereof may result in embedded input data and / or may be the output of any one of the encoder input 1078, 1084, 1088 or decoder input 1084, 1094. The data obtained by applying the token embedding, the positional embedding, the segment embedding, the column embedding, the row embedding or a combination thereof may be processed by the encoder block 1074, 1086, decoder block 1080, 1090, encoder output 1076, decoder output 1092, 1082.

[0194] FIG. 13 illustrates an embodiment of input embedding.

[0195] Input data to the data-driven model, in particular to the encoder input and / or the decoder input as described in the context of FIG. 10A-C, may comprise image data. The data-driven model may be parametrized to receive image data. For processing image data as input data, the data-driven model may comprise one or more encoder blocks and / or one or more decoder blocks and / or one or more encoder outputs and / or one or more decoder outputs as described within the context of FIG. 10A-C. FIG. 13 may show an embodiment of an encoder input and / or a decoder input. When processing image data, the encoder input and / or the decoder input of the data-driven model240436

[0196] 31

[0197] may be as described within the context of FIG. 13. The encoder input and / or decoder input may comprise one or more linear projection layers 1314 for a linear projection of one or more images, preferably one or more partial images, more preferably a sequence of two or more partial images. The one or more linear projection layers 1314 may be suitable for changing the dimension of the one or more received images, preferably one or more partial images, preferably passing the one or more images, preferably partial images, through the one or more linear projection layers 1314 may result in applying image embedding, preferably partial image embedding to the one or more images and / or partial images.

[0198] Furthermore, when a sequence of two or more images and / or partial images may be received, positional embedding may be applied to the sequence, preferably by passing the sequence of one or more images and / or partial images through the one or more linear projection layers 1314. Applying positional embedding may refer to adding a positional factor. The positional factor may be different depending on the position of the image and / or the partial image within the sequence. In particular, the positional factor added to a first element of the sequence may be different to the positional factor added to a second element of the sequence. The first element of the sequence may be a first image and / or first partial image. The second element of the sequence may be a second image and / or a second partial image.

[0199] The representation of the one or more images, preferably one or more partial images, may be obtained based on the following equation:

[0200]

[0201] wherexclass is the image class embedding 1328 ,XNpis the n-th image, in particular partial image in the sequence,zo is the representation of the one or more images, preferably one or more partial images, (H,W) are the resolution of the image, in particular the image the partial images are generated on, C is the number of channels associated with the one or more image, in particular the one or more partial images and D is the dimension of the representation of the one or more images, preferably one or more partial images. Applying the partial image embedding may refer to forming the product ofXNpwith E above-described equation. Applying the positional embedding may refer to adding the factor pos according to the above-described equation.

[0202] By doing so, text-based data, numerical data, tabular data, image data or the like may be processed by one data-driven model.

[0203] The present disclosure has been described in conjunction with preferred embodiments and examples as well. However, other variations can be understood and effected by those persons skilled in the art and practicing the claimed subject-matter, from the studies of the drawings, this disclosure and the claims. Notably, in particular, the any steps presented can be performed in any order, i.e. the present disclosure is not limited to a specific order of these steps. Moreover, it is also not required that the different steps are performed at a certain place or at one240436

[0204] 32

[0205] node of a distributed system, i.e. each of the steps may be performed at different nodes using different equipment / data processing.

[0206] As used herein ..determining" also includes ..initiating or causing to determine", “generating" also includes ..initiating and / or causing to generate" and “providing” also includes “initiating or causing to determine, generate, select, send and / or receive”. “Initiating or causing to perform an action” includes any processing signal that triggers a computing node or device to perform the respective action.

[0207] In the claims as well as in the description the word “comprising” or “including” or similar wording does not exclude other elements or steps and shall not be construed limiting to the elements or steps lined out. The indefinite article “a” or “an” does not exclude a plurality. A single element or other unit may fulfill the functions of several entities or items recited in the claims. The mere fact that certain measures are recited in the mutual different dependent claims does not indicate that a combination of these measures cannot be used in an advantageous implementation or further elements may be included.

[0208] Providing in the scope of this disclosure may include any interface configured to provide data. This may include an application programming interface, a human-machine interface such as a display and / or a software module interface. Providing may include communication of data or submission of data to the interface, in particular display to a user or use of the data by the receiving entity.

[0209] Various units, circuits, entities, nodes or other computing components may be described as “configured to” perform a task or tasks. Configured to shall recite structure meaning “having circuitry that” performs the task or tasks on operation. The units, circuits, entities, nodes or other computing components can be configured to perform the task even when the unit / circuit / component is not operating. The units, circuits, entities, nodes or other computing components that form the structure corresponding to “configured to” may include hardware circuits and / or memory storing program instructions executable to implement the operation. The units, circuits, entities, nodes or other computing components may be described as performing a task or tasks, for convenience in the description. Such descriptions shall be interpreted as including the phrase “configured to.” Any recitation of “configured to” is expressly intended not to invoke 35 U.S.C. § 112(f) interpretation.

[0210] In general, the methods, apparatuses, systems, computer elements, nodes or other computing components described herein may include memory, software components and hardware components. The memory can include volatile memory such as static or dynamic random-access memory and / or nonvolatile memory such as optical or magnetic disk storage, flash memory, programmable read-only memories, etc. The hardware components may include any combination of combinatorial logic circuitry, clocked storage devices such as flops, registers, latches,240436

[0211] 33

[0212] etc., finite state machines, memory such as static random-access memory or embedded dynamic random-access memory, custom designed circuitry, programmable logic arrays, etc.

[0213] Any disclosure and embodiments described herein relate to the methods, the systems, devices, the computer program element lined out above and vice versa. Advantageously, the benefits provided by any of the embodiments and examples equally apply to all other embodiments and examples and vice versa.

[0214] All terms and definitions used herein are understood broadly and have their general meaning.

[0215] The present disclosure has been described in conjunction with preferred embodiments and examples as well. However, other variations can be understood and effected by those persons skilled in the art and practicing the claimed invention, from the studies of the drawings, this disclosure and the claims. Notably, in particular, the any steps presented can be performed in any order, i.e. the present invention is not limited to a specific order of these steps. Moreover, it is also not required that the different steps are performed at a certain place or at one node of a distributed system, i.e. each of the steps may be performed at different nodes using different equipment / data processing.

[0216] As used herein ..determining" also includes ..initiating or causing to determine", “generating" also includes ..initiating and / or causing to generate" and “providing” also includes “initiating or causing to determine, generate, select, send and / or receive”. “Initiating or causing to perform an action” includes any processing signal that triggers a computing node or device to perform the respective action.

[0217] In the claims as well as in the description the word “comprising” does not exclude other elements or steps. The indefinite article “a” or “an” and the definite article “the” does not exclude a plurality. In particular, indefinite article “a” or “an” may be replaced with one or more and the definite article “the” may be replaced with the one or more. A single element or other unit may fulfill the functions of several entities or items recited in the claims. The mere fact that certain measures are recited in the mutual different dependent claims does not indicate that a combination of these measures cannot be used in an advantageous implementation.

[0218] Any disclosure and embodiments described herein relate to the methods, the systems, devices, the computer program element lined out above and vice versa. Advantageously, the benefits provided by any of the embodiments and examples equally apply to all other embodiments and examples and vice versa.

Claims

24043634CLAIMSWhat is claimed is:

1. A method for controlling and / or monitoring a target chemical plant, the method comprising:obtaining, in particular receiving, malfunctional data associated with a difference between a current operation and / or configuration associated with a target chemical plant and a target operation and / or configuration associated with the target chemical plant,obtaining, in particular receiving, a plurality of historical plant data sets associated with a plurality of differences between historical operations and / or configurations and target operations and / or configurations associated with a plurality of chemical plants and a plurality of causes related to the plurality of differences and / or a plurality of measures for reducing an effect of the differences onto the operation of the target chemical plant,selecting at least a part of the historical plant data sets by matching at least a part of the malfunctional data and the plurality of historical plant data sets,providing one or more digital representation(s) of at least the selected part of the historical plant data sets and a digital representation of the malfunctional data,selecting a subgroup of at least the selected part of the historical plant data sets by matching the one or more digital representation(s) of at least the part of the historical plant data sets and the digital representation of the malfunctional data,providing, in particular via a user interface, the one or more measure(s) and / or one or more cause(s) associated with the selected subgroup of the historical plant data sets for controlling and / or monitoring the operation and / or the configuration of the target chemical plant.

2. The method of claim 1 , wherein providing the malfunctional data is triggered by detecting a difference between the current operation and / or configuration associated with the target chemical plant and the target operation and / or configuration associated with the target chemical plant by one or more sensor(s) associated with the target chemical plant.

3. The method of claim 1 or 2, wherein the one or more measure(s) and / or the one or more cause(s) are provided via a user interface and / or displayed for implementing the one or more measure(s) and / or reducing the effect of the one or more cause(s) on the operation of the target chemical plant.

4. The method of any one of claims 1 to 3, wherein the malfunctional data and / or the one or more measure(s) and / or the one or more cause(s) are human-interpretable and / or may comprise natural language.240436355. The method of any one of claims 1 to 4, wherein the malfunctional data comprises and / or is indicative of one or more intermediate cause(s), and wherein the plurality of causes associated with the historical plant data sets comprises a plurality of root causes and optionally one or more intermediate cause(s) associated with said plurality of root causes, wherein the intermediate cause is a result of the plurality of root causes.

6. The method of any one of claims 1 to 5, wherein the malfunctional data includes sensor data collected by one or more sensor(s) for monitoring the target chemical plant.

7. The method of any one of claims 1 to 6, wherein providing the one or more measure(s) and / or cause(s) includes providing the selected subgroup of the historical plant data sets.

8. The method of any one of claims 1 to 7, wherein selecting at least a part of the historical plant data sets by matching at least a part of the malfunctional data and the plurality of historical plant data sets comprises selecting at least a part of the historical plant data sets by matching a plant type associated with the target chemical plant and plant types associated with the plurality of chemical plants.

9. The method of any one of claims 1 to 8, wherein selecting at least a part of the historical plant data sets by matching at least a part of the malfunctional data and the plurality of historical plant data sets comprises selecting at least a part of the historical plant data sets by matching one or more element(s) associated with at least a part of the malfunctional data and one or more element(s) per of historical plant data set.

10. The method of any one of claims 1 to 9, wherein selecting at least the part of the historical plant data sets and / or selecting the subgroup of at least the selected part of the historical plant data sets based on a digital representation of at least the part of the historical plant data sets and a digital representation of the malfunctional data includes matching the digital representation of at least the part of the historical plant data sets and the digital representation of the malfunctional data.

11. The method of any one of claims 1 to 10, wherein selecting the subgroup of at least the selected part of the historical plant data sets based on a digital representation of at least the part of the historical plant data sets and a digital representation of the malfunctional data includes processing of the digital representations by a data-driven model, wherein the data-driven model is suitable for matching the historical plant data sets and the malfunctional data based on the digital representations of the malfunctional data and the historical plant data sets.

12. The method of any one of claims 1 to 11 , further comprising determining one or more cause cluster(s) comprising one or more historical plant data set(s) from the subgroup of historical plant data sets24043636associated with at least one cause related to the plurality of the differences, and wherein providing the one or more cause(s) comprises providing one or more cause(s) per cause cluster.

13. The method of any one of claims 1 to 12,further comprising determining one or more cause cluster(s) comprising one or more historical plant data set(s) from the subgroup of historical plant data sets associated with at least one cause, and determining one or more measure cluster(s) per cause cluster comprising one or more historical plant data set(s) from the subgroup of historical plant data sets associated with at least one measure,and wherein providing the one or more cause(s) comprises providing the at least one cause per cause cluster and wherein providing the one or more measure(s) comprises providing the at least one measure per measure cluster.

14. An apparatus for controlling and / or monitoring one or more target chemical plant(s), the apparatus comprising:a processor configured for performing any one of the methods according to any one of claims 1 to 9.

15. Use of one or more measure(s) and / or cause(s) as obtained by any one of the any one of claims 1 to 13 for controlling and / or monitoring one or more target chemical plant(s).

Images