Method for operating an industrial environment with at least two different systems by means of an electronic computing device, computer program product, computer-readable storage medium and electronic computing device

Abstract

The invention relates to a method for operating an industrial environment (10) with at least two different systems (12, 14) by means of an electronic computing device (16), comprising the steps: - Providing a machine learning algorithm (24) for semantically comparing a first parameter (20) of a first installation (12) of the industrial environment (10) with at least one second parameter (22) of a second installation (14) of the industrial environment (10); - providing a low-level adaptation algorithm (26) for the machine learning algorithm (24); - Semantically comparing the first parameter (20) with the second parameter (22); - adapting the low-level adaptation algorithm (26) depending on the semantic comparison of the first parameter (20) with the second parameter (22); - operating the industrial environment (10) in accordance with the machine learning algorithm (24) and the adapted low-level adaptation algorithm (26). Furthermore, the invention relates to a computer program product, a computer-readable storage medium, and an electronic computing device (16).

Description

The invention relates to a method for operating an industrial environment having at least two different installations by means of an electronic computing device according to the applicable patent claim 1.As the number of interacting enterprises increases, as well as a large number of suppliers of complex assemblies, the need for data interoperability is one of the greatest challenges for, for example, plant designers. Requirements for interoperability are the digital and actual representation of the system components by a uniform information model and the guarantee of semantic interoperability of the systems. It describes a joint understanding of the system components about the meaning of the exchanged information. An example implementation of semantic interoperability by implementing homogeneous semantics in the form of standardized vocabularies. Existing standards for defining terms and semantics, such as ECLASS, may be used for this purpose. However, compliance with these standards is not absolutely necessary in connection with the Asset Administration Shell (AAS), which is an essential component of digital twins. The semantic diversity is represented in AAS information models by inherited so-called sub-model elements.For example, although standards are to be set by ECLASS, it is expected that due to the variety of manufacturers, products and process participants, heterogeneity of the name conventions will remain. It is therefore necessary to be able to map different vocabularies to a standardized target semantics. In this context, mapping by means of linked data is associated with a high initial outlay. In addition to the time required for creating a naming catalog, in particular the translation into different languages, this also includes the maintenance of the catalog and the possible error sources by the manual transmission of information. Particularly for changing suppliers to be integrated into the AAS system, the manual creation of an attribute-level mapping list is not relative.It is an object of the present invention to provide a method, a computer program product, a computer-readable storage medium and an electronic computing device, by means of which an automated mapping process for different parameters of different installations can be realized.This object is achieved by a method, a computer program product, a computer-readable storage medium and an electronic computing device according to the independent patent claims. Advantageous embodiments are specified in the dependent claims.One aspect of the invention relates to a method for operating an industrial environment having at least two different installations by means of an electronic computing device. A machine learning algorithm is provided for semantically comparing a first parameter of a first installation of the industrial environment with at least one second parameter of the second installation of the industrial environment by means of the electronic computing device. A low-level adaptation algorithm is provided for the machine learning algorithm by means of the electronic computing device. The first parameter is then semantically compared with the second parameter by means of the machine learning algorithm. The low-level adaptation algorithm is adapted as a function of the semantic comparison of the first parameter with the second parameter by means of the electronic computing device. The industrial environment is operated by means of the electronic computing device as a function of the machine learning model and the adapted low-level adaptation algorithm.In particular, an automated mapping can thus be realized, which reduces the manual effort and can be provided with the aid of models from the field of natural speech processing, for example. Within the scope of the present invention, for example, the use of Large Language Models (LLM) can be used. Relevant information is based on various, predefined name combinations, for example, manufacturer-specific, bsDD or the like. This not only allows semantic matching of the properties in a single submodel, but also provides input for the transformation of units of measure, which refers to one of the main part problems. Overall, the need for a name catalog is thus eliminated.In particular, the invention thus provides for determining the similarity of two different sub-model elements according to the large language models, a model that can provide a similarity score is needed. For example, the models of SBERT can be used for this purpose. It relates to a so-called Siamessian network which contains two or more identical sub-networks, the models usually having exactly the same weights. Its goal is to calculate similarity values of the inputs. Each BERT outputs pooled set embeds. Pooling is a technique of generalization of features in a network, and in this case, mean pooling works by averaging feature groups in the BERT model. After pooling there are now two embeds. One for submodel element A and one for submodel element B. When the model is trained, the model concatenates the two embeddings, which then pass through a softmax classifier and are trained with a softmax loss function. In inference--or when the model begins prediction--the two embeds are compared, for example, by a cosine similarity function which outputs a similarity value for the two phrases or words. In contrast to semantic matching with sentence transformers, when large language models are used, an unknown sentence is fed in together with a sentence of the target vocabulary of a model. To do this, the names and descriptions of the phrases are integrated into a phrase structure, the prompt. This prompt is placed in embeddments to make an input processable for the large language model.Each word of the input string is transferred into a numerical representation, in particular a vector, describing the input designation in a multi-dimensional feature space. These embeddings are then made available to the large language model.The low-rank adaptation algorithm is in particular a so-called LoRA model, in particular with so-called low-rank adaptation of large-language models. In this case, it is provided in particular that matrices are changed, for example, while the actually machine learning algorithm remains unchanged. Thus, basic models can be provided as a machine learning algorithm and an adaptation of the low-level adaptation algorithm can be carried out without influencing the weights of the machine learning algorithm, for example. In particular in the case of a very large machine learning algorithm, this has the advantage that an appropriate adaptation does not have to be carried out here, since the low-ranking adaptation algorithm is merely adapted "in parallel" to the machine learning algorithm.In particular, in contrast to the prior art, the solution developed here is therefore not aimed at a complete fine tuning or a re-learning of the existing learning models, in particular of the existing machine learning algorithm. Rather, small adaptations to the underlying neural network, in particular the low-level adaptation algorithm, or the integration of existing methods such as parameter efficient finding (PEFT) are provided as a lightweight parameterization of the results. In addition, the approach is not based on models with a small set of parameters, but on the use of established multilingual large language models. Another distinguishing feature is the continuous update cycle using human feedback reinforcement learning (RLHF) aimed at continuously improving the solution proposed by the Large Language Model.According to an advantageous embodiment, the machine learning algorithm is provided as a large language model. The large language model is in particular the large language model. In other words, a large language model can be provided, whereby simplified matching of the parameters can be performed. Thus, on the basis of the vocabulary of the large speech model, an improved matching or mapping of the parameters can already be carried out. In order to perform only fine tuning, the low-order matching algorithm is used in parallel with the large speech model. Thus, the large language model having a large vocabulary can be used and fine tuning to the specific industrial environment based on the low-level matching algorithm can be realized at the same time.A further advantageous embodiment provides that at least one weighting is adapted within the low-level adaptation algorithm. In particular, the low-level adaptation algorithm can be provided, for example, in the form of a neural network. Corresponding weights of the neural network can then be adapted or adapted in order to carry out the adaptation in turn. Thus, an adaptation of the low-level adaptation algorithm can be realized in a simple manner.It is likewise advantageous if the low-level adaptation algorithm is provided with at least two adaptable matrices. In particular, the at least two adaptable matrices can then in turn be correspondingly multiplied with one another in order to obtain a higher-ranking matrix. The higher-rank matrix can in turn be used in particular for the corresponding weights within the lower-rank adaptation model.It is furthermore advantageous if a low-order adaptation algorithm is provided with at least one encoder device and / or one decoder device. In particular, a parameter-efficient fine tuning technique classified as reparametrization can thus be provided. In this case, for example, the input request can be converted as into tokens, which are then converted into embedding vectors and forwarded to the encoder and / or decoder parts of the transformer. Thus, a reliable low-level adaptation of the large language model can be provided.A further advantageous embodiment provides that the low-level adaptation algorithm is provided with at least one self-observation network and one feed-forward network. In particular, these two components can be used with two types of neural networks. In particular, networks with self-observation and feed-forward networks can be used. The weights of these nets are learned during pretraining. After the embedding vectors of the inputs have been created, they are fed into the self-attention layers, a series of weights being applied to calculate the attention values. During the full fine tuning, each parameter in these layers is updated. LoRA is a strategy that reduces the number of parameters to train during fine tuning by freezing all the original model parameters and then injecting a pair of edge decomposition matrices adjacent to the original weights. The dimensions of the smaller matrices are determined so that their product is a matrix with the same dimensions as the weights they change. These parameters are to be adapted depending on the specific large language model to be applied. The original weights of the large language model are then frozen and the small matrices are trained using the supervised learning method.A further advantageous embodiment provides that the machine learning algorithm is provided as a predefined machine learning algorithm. In the present case, a predefined machine learning algorithm is to be understood in particular as a basic machine learning algorithm. In other words, already prefabricated machine learning algorithms can be used to be able to carry out a corresponding operation of the industrial environment. A specific adaptation of the predefined machine learning algorithm is not necessary. The specific implementation takes place in particular via the low-level adaptation algorithm. Established machine learning algorithms can thus already be used, as a result of which improved operation of the industrial environment can be realized.It is likewise advantageous if, in addition, the low-level adaptation algorithm is adapted on the basis of optimization of results in reinforcement learning with human feedback. In this case, this method is based in particular on the fact that the results supplied by the large language model and the LoRA are to be improved further and the corresponding semantics are to be mapped even more accurately. It should also be noted in this method that, for example, only the parameters of the LoRA concept are adapted and no transposing of the large language model itself takes place. The practical application can be outlined, for example, in such a way that two submodels of the AAS are merged. These consist in each case of a set of attributes with partly matching attribute names. The large language model plus the LoRA model generates a probability matrix. Sometimes there are direct matches, such as in the case of the terms security or capacity. This then results in an adjustment of the weights of LoRA, which is necessary. The user resolves the conflicts and correctly assigns the labels. Then, a process is initiated in which a model is compared with the old LoRA values and a model is compared with the new LoRA values to optimize them. Thus, reinforcement learning may be provided so that operation of the industrial environment may be realized in more detail and more specifically, during operation.It is likewise advantageous if the optimization is carried out on the basis of a mean square error of the human feedback. In particular, the mean square error (MSE) can thus be used as an error measure to measure the deviation from the user definition. The adjustments to the LoRA weights are based on the mean square error and iteratively adjusted to approximate the ideal result. If the result of the existing model cannot be exceeded, the existing model remains in use, otherwise it is overwritten by the weight update.Furthermore, it has proven advantageous if the at least two parameters are compared by means of a cosine similarity function. In particular, a similarity value for the two phrases or words can be output on the basis of the cosine similarity function. Thus, a comparison can be realized in a simple manner and nevertheless reliably.Furthermore, it has proven advantageous if the industrial environment is converted into a digital twin model of the industrial environment, wherein the at least two installations are provided in the digital twin model on the basis of the semantic comparison. In particular, it is thus possible, for example, to provide a display for a user of the electronic computing device or of the industrial environment. The systems can then be represented accordingly in the digital twin model. For example, corresponding parameters, in particular the detected parameters, can be displayed accordingly. On the basis of the result of the electronic computing device, values can be matched or mapped accordingly, so that it is easily possible for a user of the electronic computing device to see which parameters are currently provided by the installations. For example, if a watt number should come from a first installation as the parameter value and a power number in kilowatts should be specified by a second installation, this can now be indicated on the basis of the mapping in such a way that a user can reliably understand which actual powers are being queried in a common display of the correct dimension within the installations.It is furthermore advantageous if at least the low-level adaptation algorithm is provided as a neural network. Here and in the following, an artificial neural network can be understood as software code which is stored on a computer-readable storage medium and represents one or more networked artificial neurons or can simulate their function. The software code can also contain a plurality of software code components, which can have different functions, for example. In particular, an artificial neural network may implement a nonlinear model or algorithm that maps an input to an output, where the input is given by an input feature vector or sequence and the output may include, for example, an output category for a classification task, one or more predicted values, or sequence.The presented method is in particular a computer-implemented method. Therefore, a further aspect of the invention relates to a computer program product having program code means which cause an electronic computing device to carry out a method according to the preceding aspect when the program code means are executed by the electronic computing device.Furthermore, the invention therefore also relates to a computer-readable storage medium having at least the computer program product.Yet another aspect of the invention relates to an electronic computing device for operating an industrial environment having at least two different installations, having at least one machine learning algorithm and a low-level adaptation algorithm, wherein the electronic computing device is designed to carry out a method according to the preceding aspect. In particular, the method is carried out by means of the electronic computing device.Furthermore, the invention also relates to an industrial environment having at least two different installations and the electronic computing device according to the preceding aspect.Advantageous embodiments of the method are to be regarded as advantageous embodiments of the computer program product, the computer-readable storage medium, the electronic computing device and the industrial environment. For this purpose, the electronic computing device and the industrial environment have subject-matter features in order to be able to carry out corresponding method steps.A computing unit / electronic computing device can be understood in particular as a data processing device which contains a processing circuit. The computing unit can therefore process data in particular for carrying out computing operations. This also includes operations to perform indexed accesses to a data structure, for example a look-up table (LUT).The computing unit can in particular contain one or more computers, one or more microcontrollers and / or one or more integrated circuits, for example one or more application-specific integrated circuits, ASICs (application-specific integrated circuit), one or more field programmable gate arrays, FPGAs, and / or one or more single-chip systems, SoCs (system on a chip). The computing unit may also contain one or more processors, for example one or more microprocessors, one or more central processing units, CPUs (central processing units), one or more graphics processing units, GPUs (graphics processing units) and / or one or more signal processors, in particular one or more digital signal processors, DSPs. The computing unit may also include a physical or virtual group of computers or other of the aforementioned units.In various exemplary embodiments, the computing unit contains one or more hardware and / or software interfaces and / or one or more memory units.A memory unit can be used as volatile data memory, for example as dynamic random access memory, DRAM (dynamic random access memory) or static random access memory, SRAM (static random access memory), or as nonvolatile data memory, for example as read-only memory, ROM (read-only memory), as programmable read-only memory, PROM (programmable read-only memory), as erasable programmable read-only memory, EPROM (erasable programmable read-only memory), as electrically erasable programmable read-only memory, EEPROM (electrically erasable programmable read-only memory), as flash memory or flash EEPROM, as ferroelectric random access memory, FRAM (ferroelectric random access memory), as magnetoresistive random access memory, MRAM (magnetoresistive random access memory) or as phase-change random access memory, PCRAM (phase-change random access memory).For use cases or application situations which can arise in the method and which are not explicitly described here, provision can be made for an error message and / or a request for inputting a user feedback to be output and / or for a default setting and / or a predetermined initial state to be set according to the method.Regardless of the grammatical sex of a certain term, individuals with male, female or other sex identity are included.Further features and combinations of features of the invention are evident from the figures and their description and from the claims. In particular, further embodiments of the invention need not necessarily include all features of any of the claims. Further embodiments of the inventions may comprise features or combinations of features not mentioned in the claims.The following are shown: FIG. 1 is a schematic block diagram according to an embodiment of an industrial environment; and FIG. 2 is a schematic block diagram according to an embodiment of an electronic computing device.The invention is explained in more detail below with reference to specific exemplary embodiments and associated schematic drawings. In the figures, identical or functionally identical elements can be provided with the same reference numerals. The description of identical or functionally identical elements may not necessarily be repeated with respect to different figures.FIG. 1 shows a schematic block diagram according to an embodiment of an industrial environment 10. the industrial environment 10 comprises a first installation 12 and a second installation 14 different from the first installation 12. The industrial environment 10 comprises at least one electronic computing device 16. The electronic computing device 16 can have a display device 18, for example. Results of the electronic computing device 16 can be displayed on the display device 18, for example.In particular, a first parameter 20 can be transmitted from the first installation 12 to the electronic computing device 16. A second parameter 22 can be transmitted from the second installation 14 to the electronic computing device 16.FIG. 2 shows a schematic block diagram according to one specific embodiment of electronic computing device 16. electronic computing device 16 has at least one machine learning algorithm 24 and one low-level adaptation algorithm 26 in the present exemplary embodiment.Further, the electronic computing device 16 may include an input layer 28 and an output layer 30.According to one embodiment of the method, it is provided that the machine learning algorithm 24 is provided for semantically comparing the first parameter 20 with the second parameter 22. For this purpose, the low-level adaptation algorithm 26 is also provided for the machine learning algorithm 24, in particular in parallel, as shown in the present case. Then, the first parameter 20 is semantically compared with the second parameter 22 by means of the machine learning algorithm 24, and the adaptation of the low-level adaptation algorithm 26 is then carried out again as a function of the semantic comparison of the first parameter 20 with the second parameter 22 by means of the electronic computing device 16 and the operation of the industrial environment 10 as a function of the machine learning algorithm 24 and the adapted low-level adaptation algorithm 26.In this case, it is provided in particular that the machine learning algorithm 24 is provided as a large speech model. Furthermore, it can be provided that at least one weighting is adapted within the low-level adaptation algorithm 26. The low-level adaptation algorithm 26 can be provided with at least two adaptable matrices. Furthermore, the low-level adaptation algorithm 26 can be provided with at least one encoder device 34 and one decoder device 36. In this case, as shown in FIG. 2, the low-level adaptation algorithm 26 is preferably provided as a neural network. In this case, the low-level adaptation algorithm 26 can be provided with at least one self-monitoring network and a feed forward network.The machine learning algorithm 24 may be provided, for example, as a large language model. Furthermore, it can be provided that the low-level adaptation algorithm 26 is additionally adapted on the basis of an optimization of results in a reinforcement learning 32 with human feedback.In particular, the figures thus show in order to determine the similarity of two different sub-model elements (SE), for example a model which can provide a similarity assessment is required. One of the most commonly used examples is SBERT. It relates to a so-called Siamesian network which contains two (or more) identical sub-networks, the models generally having exactly the same weights. Its goal is to calculate similarity values of the inputs.Each BERT outputs pooled set embeds. Pooling is a technique for generalization of features in a network, and in this case, mean pooling works by averaging feature groups in the BERT model. After pooling, there are now two embeds: one for sub-model element A and one for sub-model element B. When the model is trained, the model concatenates the two embeds, which are then passed through a softmax classifier and trained with a softmax loss function. In inference--or when the model begins prediction--the two embeds are then compared to a cosine similarity function which outputs a similarity value for the two phrases or words. In contrast to semantic matching with sentence transformers, when using LLMs (Large Language Model), an unknown SE is fed into the model together with an SE of the target vocabulary. For this purpose, the names and descriptions of the SEs are integrated into a sentence structure, the prompt. This prompt is transferred to embeds to make the input processable for the LLM.Each word of the input string is transferred into a numerical representation (vector) describing the input string in a multi-dimensional feature space. These embeds are then provided to the LLM.LoRA is a parameter-efficient fine tuning technique classified as reparametrization. The input request is converted into tokens which are then converted into embedding vectors and passed to the encoder and / or decoder parts of the transformer. Two types of neural networks are used in these two components: self-observation networks and feedforward networks. The weights of these nets are learned during pretraining. After the embedding vectors of the inputs have been created, they are fed into the self-attention layers where a series of weights are applied to calculate the attention values. During the full fine tuning, each parameter in these layers is updated. LoRA is a strategy that reduces the number of parameters to train during fine tuning by freezing all the original model parameters and then injecting a pair of rank decomposition matrices adjacent the original weights. The dimensions of the smaller matrices are determined so that their product is a matrix with the same dimensions as the weights they change. This parameter is to be adjusted depending on the specific LLM to be applied. The original weights of the LLM are then frozen and the smaller matrices are trained using the supervised learning method.For inference, the two low rank matrices are multiplied together to produce a matrix having the same dimensions as the frozen weights. These are then added to the original weights and replaced in the model with these updated values. Now there is a fine tuned LoRA model that can perform specific semantic matches. Since most parameters of LLMs are in the attention layers, the greatest savings in trainable parameters can be achieved when LoRA is applied to these weight matrices. LoRA is applied for each matching of two semantics to adapt to the specific application task during the training process.Normally, the use of an LLM starts with a compute-intensive training process of the LLM itself. Because this approach aims to reduce the computational resources rather than training an LLM for each semantic matching, the use of already established LLMs is proposed. This could be a generalized LLM such as GPT or already specialized models for technical data.In the training process, already known semantic matches are used to initialize the values of the matrices as part of the LoRA approach. This process uses matching semantic pairs as part of a probability matrix.In this matrix, each value indicates the semantic similarity between the attributes in the rows and columns, e.g., 1.00 representing a perfect match. The matrix is the result of a pairwise query on the LLM if the two words or descriptions are similar, giving a similarity value between zero and one.The similarity score itself refers to cosine similarity, analogous to the SBERT approach. As a result, the LoRA weights are adjusted to achieve the result of the real semantic matches as well as possible.Since there may still be false semantic assignments, new semantics to be assigned, or other problems in productive use, the human feedback reinforcement learning 32 method is used to achieve further improvements.This method is based on the result supplied by LLM+LoRA being further improved and the corresponding semantics being mapped even more accurately. It should also be noted in this method that only the parameters of the LoRA concept are adapted and no transposing of the LLM itself takes place. The practical application can be outlined as follows: two submodels of the AAS (Asset Administration Shell), i.e. of the industrial environment 10, are to be merged. These consist in each case of a set of attributes with partly matching attribute names. The LLM+LoRA generates a probability matrix as shown below. Sometimes there are direct matches, as in the case of security or capacity. The corresponding assignment of the terms would be as follows:Efficiency >LeistungSecurity->SReliability->Capacity->resliabilities already due to 1:1 relationship / higher probability of another wordCost->efficiency is already achieved by a 1:1 relationship / higher probability for other wordsLongevity->As a result, adjustment of the weights of LoRAis necessary. The user resolves the conflicts and correctly assigns the labels. A process is then initiated in which a model is compared with the old LoRA values and a model with new LoRA values in order to optimize them.The mean square error (MSE) is used as an error measure to measure the deviation from the user definition. The adjustments to the LoRA weights are iteratively adjusted to approximate the ideal result. If the result of the existing model cannot be exceeded, the existing model remains in use, otherwise it is overwritten by the weight updates.In the case of a new matching between new semantics, the basic model must be used, which must then be adapted to the corresponding circumstances in the corresponding sub-method. The LoRA values are only adapted here after the first user adaptation, since the semantic matching of the LLM without LoRA may also be sufficient. The LoRA weights are therefore initialized to have no influence.List of reference characters10 Industrial Environment 12 First Installation 14 Second Installation 16 Electronic computing device 18 Display device 20 First parameter 22 Second parameter 24 Machine learning algorithm 26 Low-order adaptation algorithm 28 Input layer 30 Output layer 32 Reinforcement learning 34 Encoder device 36 Decoder device

Claims

Method for operating an industrial environment (10) having at least two different installations (12, 14) by means of an electronic computing device (16), having the steps: - providing a machine learning algorithm (24) for semantically comparing a first parameter (20) of a first installation (12) of the industrial environment (10) with at least one second parameter (22) of a second installation (14) of the industrial environment (10) by means of the electronic computing device (16); - providing a low-level adaptation algorithm (26) for the machine learning algorithm (24) by means of the electronic computing device (16); - semantically comparing the first parameter (20) with the second parameter (22) by means of the machine learning algorithm (24); - adapting the low-level adaptation algorithm (26) on the basis of the semantic comparison of the first parameter (20) with the second parameter (22) by means of the electronic computing device (16); operating the industrial environment (10) as a function of the machine learning algorithm (24) and the adapted low-level adaptation algorithm (26) by means of the electronic computing device (16).The method of claim 1, characterized in that the machine learning algorithm (24) is provided as a large language model.Method according to Claim 1 or 2, characterized in that at least one weighting is adapted within the low-level adaptation algorithm (26).Method according to one of the preceding claims, characterized in that the low-level adaptation algorithm (26) is provided with at least two adaptable matrices.Method according to one of the preceding claims, characterized in that the low-level adaptation algorithm (26) is provided with at least one encoder device (34) and one decoder device (36).Method according to one of the preceding claims, characterized in that the low-level adaptation algorithm (26) is provided with at least one self-observation network and one feedforward network.Method according to one of the preceding claims, characterized in that the machine learning algorithm (24) is provided as a predefined machine learning algorithm (24).Method according to one of the preceding claims, characterized in that the low-level adaptation algorithm (26) is additionally adapted on the basis of an optimization of results in a reinforcement learning (32) with human feedback.Method according to claim 8, characterised in that the optimisation is carried out on the basis of a mean square error of the human feedback.Method according to one of the preceding claims, characterized in that the at least two parameters (20, 22) are compared by means of a cosine similarity function.Method according to one of the preceding claims, characterized in that the industrial environment (10) is transferred into a digital twin model of the industrial environment (10), wherein the at least two installations (12, 14) are provided in the digital twin model on the basis of the semantic comparison.Method according to one of the preceding claims, characterized in that at least the low-level adaptation algorithm (26) is provided as a neural network.Computer program product having program code means which cause an electronic computing device (16), when the program code means are executed by the electronic computing device (16), to carry out a method according to one of Claims 1 to 12.A computer readable storage medium comprising at least one computer program product according to claim 13.Electronic computing device (16) for operating an industrial environment (10) having at least two different installations (12, 14), having at least one machine learning algorithm (24) and a low-level adaptation algorithm (26), wherein the electronic computing device (16) is designed to carry out a method according to one of Claims 1 to 12.

Images