Monitoring of a complex process having at least two process steps

Abstract

A computer-implemented method for monitoring a complex chemical process having at least two process steps, comprising: providing monitoring data by at least one monitoring process for each one of the at least two process steps; providing aggregation data by an aggregation layer by performing summarizations and / or combinations based on all of the provided monitoring data; providing instructive data based on the aggregation data.

Description

[0001] MONITORING OF A COMPLEX PROCESS HAVING AT LEAST TWO PROCESS STEPS

[0002] TECHNICAL FIELD

[0003] The present disclosure relates to a computer-implemented method for monitoring a complex process having at least two process steps, an apparatus for monitoring a complex process having at least two process steps, a use of control data for controlling or adapting the complex process and / or at least one process step of the complex process, and a computer program element.

[0004] TECHNICAL BACKGROUND

[0005] The general background of this disclosure is the monitoring of complex processes like a chemical process having at least two process steps.

[0006] In common technical practice, the monitoring of complex processes is complicated and highly complex based on the interactions of a plurality of different, in particular fundamentally different, processes or sub-processes. Therefore, in common technical practice, the monitoring of complex processes has to be split in a plurality of sub monitoring processes which each monitors different sub-processes of the whole complex process.

[0007] It has been found that a need exists for an alternative and smarter way of monitoring complex processes. Through the inclusion of monitoring processes and aggregation layers in the monitoring of the complex process, the plurality of sub monitoring processes becomes invalid, such that the monitoring of a complex process can be made easier and less complex. Further, based on this, resources like power consumption and hardware requirements can be saved and reduced.

[0008] SUMMARY OF THE INVENTION

[0009] In one aspect of the present disclosure, a computer-implemented method for monitoring a complex process having at least two process steps is presented, comprising: providing monitoring data by at least one monitoring process for each one of the at least two process steps; providing aggregation data by an aggregation layer by performing summarizations and / or combinations based on all of the provided monitoring data; providing instructive data based on the aggregation data.

[0010] In a further aspect of the present disclosure, an apparatus for monitoring a complex process having at least two process steps is presented, the apparatus comprising: one or more computing nodes; and one or more computer-readable media having thereon computer-executable instructions that are structured such that, when executed by the one or more computing nodes, cause the apparatus to perform the following steps: providing monitoring data by at least one monitoring process for each one of the at least two process steps; providing aggregation data by an aggregation layer by performing summarizations and / or combinations based on all of the provided monitoring data; providing instructive data based on the aggregation data.

[0011] Optionally, a system for monitoring a complex process having at least two process steps is presented, comprising: a first providing unit for providing monitoring data by at least one monitoring process for each one of the at least two process steps; a second providing unit providing aggregation data by an aggregation layer by performing summarizations and / or combinations based on all of the provided monitoring data; a third providing unit providing instructive data based on the aggregation data.

[0012] In a further aspect, a use of control data for controlling or adapting the complex process and / or at least one process step of the complex process is presented.

[0013] In a further aspect, a computer element, in particular a computer program product or a computer readable medium, with instructions, which when executed on computing device(s) is configured to carry out the steps of any of the method disclosed herein in an apparatus disclosed herein is presented. Any disclosure and embodiments described herein relate to the method, the apparatus, the system, 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.

[0014] As used herein ..determining" also includes ..initiating or causing to determine, estimating, calculating, modelizing", and “providing” also includes “initiating or causing to determine, generate, select, measure, send or receive”.

[0015] The method, apparatus, and computer element disclosed herein provide an efficient, sustainable and robust way for monitoring complex processes. In particular, the efficient, sustainable, and robust way for monitoring complex processes is at least based on the inclusion of a plurality of monitoring processes and aggregation layers in the monitoring of the complex process. Therefore, the plurality of sub monitoring processes becomes invalid and unnecessary, such that the monitoring of a complex process can be made easier and less complex. Further, resources like power consumption and hardware requirements can be saved and reduced.

[0016] It is an object of the present invention to provide an efficient, sustainable and robust way for monitoring complex processes. 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 preferred embodiments of the invention.

[0017] The term complex process is to be understood broadly in the present case and represents any process, which includes at least two process steps, in particular a plurality of process steps. For instance, a complex process may be a chemical process, a continuous process, a continuous chemical process, a manufacturing process, a band-reactor process, a manufacturing process of sandwich panels, a manufacturing process of foams, a manufacturing process of duroplastic foams, and / or a manufacturing process of thermoplastic foams, a manufacturing process of polyurethane, and / or a manufacturing process of polyisocyanorate, and / or a manufacturing process of polystyrene, and / or a manufacturing process of melamine resin, but is not limited thereto. The complex process may include a plurality of sub-processes. The term monitoring process is to be understood broadly in the present case and represents any execution of an algorithm or model being configured for monitor, i.e. watching, tracking, measuring, checking and / or observing, a condition, a function, a process, and / or manufacturing process. Further, the monitoring process is configured for providing monitoring data being indicative of a result of the monitoring of the condition, the function, the process, and / or the manufacturing process. The monitoring data may be provided with a timestamp of the providing / measurement time. For instance, the monitoring process may be a machine learning algorithm, a neuronal network, a measurement, a Region based Convolutional Neural Network, an Al, i.e. artificial intelligence, model, a machine learning network with a residual neuronal network, a machine learning model, a classification model and / or a sum square of deviations algorithm, but is not limited thereto. The plurality of monitoring processes can be identical or different to each other, i.e. can monitor same or different parameters or variables. The monitoring process is executed for solely one process step of the at least two process steps of the complex process or is executed separately for a plurality of process step of the at least two process steps of the complex process, i.e. each process step is monitored by his own monitoring process.

[0018] The aggregation layer provides aggregation data by performing summarizations and / or combinations based on all of the provided monitoring data. In particular, the aggregation layer may summarize and / or combine all of the provided monitoring data and / or of provided intermediate aggregation data. That is, to provide the aggregation data, monitoring data for each one of the at least two process steps is summarized and / or combined. More particularly, the term aggregation layer is to be understood broadly in the present case and represents any algorithm and / or model for summarizing, combining and / or comparing all of the provided monitoring data and / or provided intermediate aggregation data with reference data. In other words, the aggregation layer is the final aggregation layer, which receives all provided data, i.e. monitoring data and / or intermediate aggregation data. For instance, the aggregation layer may be an algorithm or a machine-learning algorithm, but is not limited thereto. The aggregation layer may receive the provided monitoring data and / or provided intermediate aggregation data from all, i.e. a plurality of, identical or different monitoring processes and / or intermediate aggregation layers. The results of the aggregation layer are provided as data, i.e. the so- called aggregation data. Reference data may be any data being predetermined, pre-set and / or pre-defined indicating an optimal operation state. The reference data set can be extracted during normal steady-state operation provided that machines and facilities are operating in their optimum parameter window. It may be noteworthy that data is transformed stepwise from an abstract measurement date to a specific process control data set with each integration layer.

[0019] The term comparing or comparison is to be understood broadly in the present case and represents any process for matching or correlating the provided monitoring data and / or provided intermediate aggregation data with reference data.

[0020] The term instructive data is to be understood broadly in the present case and represents any data being indicative of the state of the complex process. For instance, the instructive data may include alarms, interruptions of the complex process, error notifications in the complex process and / or recommendations how to react on interruptions, alarms and / or errors. The instructive data are based on aggregation data.

[0021] The term intermediate aggregation layer is to be understood broadly in the present case and represents any algorithm and / or model for summarizing, combining and / or comparing one or a plurality of the provided monitoring data with reference data. For instance, the intermediate aggregation layer may be an algorithm or a machine-learning algorithm, but is not limited thereto. The intermediate aggregation layer may receive the provided monitoring data from one or a plurality of identical or different monitoring. The results of the intermediate aggregation layer are provided as data, i.e. the so-called intermediate aggregation data. The intermediate aggregation layer transmits the provided intermediate aggregation data to the aggregation layer for further processing. Reference data may be any data being predetermined, pre-set and / or pre-defined indicating an optimal operation state. Thus, an intermediate aggregation layer behaves similar to the before mentioned aggregation layer but splits the aggregation task into smaller, modular steps.

[0022] The term providing as used herein is to be understood broadly in the present case and represents any method for receiving, measuring, determining, generating, selecting, sending, or receiving of parameter or data. For instance, data can be provided respectively received by the internet or can be provided or changed / adapted by a manual input by the user via a user interface.

[0023] The control data as used herein is to be understood broadly in the present case and relates to any data configured to control or adapt a complex and continuous chemical process and / or at least one process step of the complex and continuous chemical process based on the instructive data. The control data are provided by a control unit and may be configured to control one or more technical means of a process / manufacturing device, but is not limited thereto. The control data may be one-dimensional, two-dimensional or three-dimensional data. The control data are based on the provided instructive data.

[0024] In an embodiment the method for monitoring a complex process having at least two process steps, the aggregation data are provided by the aggregation layer by a comparison of the provided monitoring data with reference data. Aggregation layer allow splitting the complex monitoring task of the complex process into smaller monitoring modules. Each of the small monitoring module is modular and can be developed and maintained independently. In addition, these small monitoring modules can be re-used and reconfigured with the help of the aggregation layer to monitor different complex processes.

[0025] In a further embodiment the method for monitoring a complex process having at least two process steps, the monitoring data of each one of the at least two process steps are provided by a separate monitoring process of the at least one monitoring process.

[0026] In a further embodiment the method for monitoring a complex process having at least two process steps, the monitoring data of at least two process steps of the at least two process steps are provided by a same monitoring process of the at least one monitoring process.

[0027] In a further embodiment the method for monitoring a complex process having at least two process steps, the providing of aggregation data by the aggregation layer is provided by an algorithm, in particular a machine-learning algorithm. A machine-learning algorithm may be much more versatile in terms of adoption to different complex processes or process steps. Moreover, machine-learning algorithms are in general more robust to non- process relevant environmental conditions like e.g. illumination, shadowing, colorbalance or noise.

[0028] In a further embodiment the method for monitoring a complex process having at least two process steps, the method further comprises the step of providing intermediate aggregation data by an intermediate aggregation layer based on a subset of the provided monitoring data, wherein the providing of aggregation data by the aggregation layer is based on the provided intermediate aggregation data. In particular, the providing of aggregation data by the aggregation layer is based on all of the provided intermediate aggregation data.

[0029] In a further embodiment the method for monitoring a complex process having at least two process steps, the providing of intermediate aggregation data by the aggregation layer is provided by an algorithm, in particular a machine learning algorithm.

[0030] In a further embodiment the method for monitoring a complex process having at least two process steps, the monitoring process is at least one out of a group, the group consisting of a machine learning algorithm, a neuronal network, a measurement, a Region based Convolutional Neural Network, an Al model, a machine learning network with a residual neuronal network, a machine learning model, a classification model and a sum square of deviations algorithm.

[0031] In a further embodiment the method for monitoring a complex process having at least two process steps, when the at least one monitoring process is a plurality of monitoring processes, the monitoring processes are different to each other. Alternatively, when the at least one monitoring process is a plurality of monitoring processes, the monitoring processes are identical to each other. By using monitoring processes which are different to each other, when the at least one monitoring process is a plurality of monitoring processes, a monitoring of a plurality of different process steps can be monitored in specific manners.

[0032] In a further embodiment the method for monitoring a complex process having at least two process steps, the instructive data includes at least one out of a group, the group consisting of alarms, interruptions of the complex process, errors notifications in the complex process and recommendations how to react on interruptions, alarms and errors, respectively. This instructive data allows feedback loops. One possible feedback loop is direct manual or automatic interaction with the process control system in order to keep the complex process in the optimal process window or to shift process parameters to move the complex process into its optimal window in case it has left or is starting to leave the optimal process window. Moreover, this instructive data can also be used to automatically loop back and re-train the monitoring algorithms. Finally the instructive data may even be used to automatically find more optimized production windows by means of un-attended, computer driven, e.g. brute force feedback trial and error loops.

[0033] In a further embodiment the method for monitoring a complex process having at least two process steps, the method further comprises the step of providing control data for controlling or adapting at least one process step of the complex process based in the instructive data. Alternatively or additionally, the providing control data are capable for controlling or adapting the whole complex and continuous chemical process based on the instructive data. By providing control data, an automation can be provided.

[0034] In a further embodiment the method for monitoring a complex process having at least two process steps, the complex process is at least one out of a group, the group consisting of a chemical process, a continuous process, a complex and continuous chemical process, a manufacturing process, a band-reactor process, a manufacturing process of sandwich panels, a manufacturing process of foams, a manufacturing process of polyurethane, and / or a manufacturing process of polyisocyanorate.

[0035] BRIEF DESCRIPTION OF THE DRAWINGS

[0036] In the following, the present disclosure is further described with reference to the enclosed figures:

[0037] Fig. 1 illustrates a flow diagram of a computer-implemented method for monitoring a complex process having at least two process steps; Fig. 2 illustrates an overview of the general method for monitoring a complex process having at least two process steps; ML abbreviates machine learning;

[0038] Fig. 3 illustrates a flow diagram of an exemplary process for producing an insulation panel using the computer-implemented method for monitoring a complex process; and

[0039] Fig. 4 illustrates a box diagram of a system for monitoring a complex process having at least two process steps.

[0040] DETAILED DESCRIPTION OF EMBODIMENTS

[0041] The following embodiments are mere examples for implementing the method and the system disclosed herein and shall not be considered limiting.

[0042] Fig. 1 illustrates a flow diagram of a computer-implemented method for monitoring a complex process, i.e. a chemical process, a continuous process, a manufacturing process, a band reactor, a manufacturing process of sandwich panels, a manufacturing process of foams, a manufacturing process of duroplastic foams, a manufacturing process of thermoplastic foams, a manufacturing process of polyurethane, a manufacturing process of polyisocyanorate, a manufacturing process of polystyrene, a manufacturing process of melamine resin, having at least two process steps. In the following, an exemplary order of the steps according to the present disclosure is explained. However, the provided order is not mandatory, i.e. all or several steps may be performed in a different order or simultaneously.

[0043] The method steps shown in Fig. 1 may be executed by the systems.

[0044] In a first step, monitoring data are provided by at least one monitoring process for each one of the at least two process steps. The monitoring process is a machine learning algorithm, a neuronal network, a measurement, a Region based Convolutional Neural Network, an Al model, a machine learning network with a residual neuronal network, a machine learning model, a classification model and / or a sum square of deviations algorithm. The monitoring data represent the result of the monitoring process. The monitoring data of at least one of the at least two process steps are provided by a same monitoring process of the at least one monitoring process.

[0045] In a second step, aggregation data are provided by an aggregation layer by performing summarizations and / or combinations based on all of the provided monitoring data, in particular the aggregation data are further provided by a comparison of the provided monitoring data with reference data. The providing of aggregation data by the aggregation layer is provided by a machine-learning algorithm.

[0046] In a third step, instructive data are provided based on the aggregation data. The instructive data includes alarms, interruptions of the complex process, errors in the complex process and / or recommendations how to react on interruptions, alarms and / or errors.

[0047] Optionally, the method further comprises the step of providing intermediate aggregation data by an intermediate aggregation layer based on a subset of the provided monitoring data. The providing of aggregation data by the aggregation layer is based on all of the provided intermediate aggregation data. The providing of intermediate aggregation data is provided by the aggregation layer by a machine learning algorithm.

[0048] Optionally, the method further comprises the step of providing control data for controlling or adapting the complex process and / or at least one process step of the complex process.

[0049] Fig. 2 illustrates an overview of the general method for monitoring a complex process having at least two process steps.

[0050] A complex process includes a plurality of process steps, in particular sub-steps. Each of the process steps is monitored by another monitoring process and / or by the same monitoring process. The monitoring process is a machine learning, ML algorithm, or other measurements, which are provided by e.g. a sensor. The machine learning algorithms can be identical or different. The other measurements can be identical or different. The results of the monitoring processes are monitoring data. The monitoring data are provided to at least one intermediate aggregation layer. The intermediate aggregation layer is a machine learning algorithm. The intermediate aggregation layer provides intermediate aggregation data and provided these intermediate aggregation data to an aggregation layer. The aggregation layer includes also a machine learning algorithm for providing aggregation data. Based on the aggregation data a monitoring, a controlling etc. can be provided.

[0051] Fig. 3 illustrates a flow diagram of an exemplary process for producing an insulation panel using the computer-implemented method for monitoring a complex process. The process provides a process and quality control of complex industrial chemical processes by means of multiple (n>2), parallel, independently operating, simple neuronal networks.

[0052] The process included the following steps for an insulation panel production with a solid foam core. Especially, sandwich panel production process: Insulation foam core sandwiched on both laminar sides with coated metal cover, i.e. , top and bottom surface is covered by metal sheet. Foam is produced directly in line by polyurethane or polyisocyanurate reaction.

[0053] In a step 1 , a coated metal cover (coil) is preprocessed. During preprocessing the following sub-steps are executed: profiling of metal covering (surface area and edge connection profile), cleaning of metal surface (mechanically, corona, ... ), and heating to prepare for ideal Pll reaction conditions. During preprocessing, the surveillance is to detect any defects, voids, scratches, impurities, flaws in or onto the surface of the (coated) metal sheet. Ideally, this surface should be homogeneous and fully cleaned. For detecting any defects, voids, scratches, impurities, flaws in or onto the surface of the (coated) metal sheet, optical cameras are used to cover full width of metal sheet, provide video / image data from multiple cameras is stitched together to result in full image of full width of metal sheet. Additionally, a Faster R-CNN (Region based Convolutional Neural Network) network is trained to detect any defects, voids, scratches, impurities, flaws in or onto the surface of the (coated) metal sheet that is different from the clean (coated) metal sheet. Ideal output of the Faster R-CNN is zero, i.e. nothing detected. Any output of the Faster R-CNN being different from zero indicates a potential problem. The Faster R-CNN, i.e. Al model, is trained with (human supervised) OK-data from a normally running solution. Defect data has been generated artificially by intentionally adding scratches, liquid and solid impurities onto the metal surface during production stops. For an alternate complex process we use YOLO Object Detection Algorithm for detect the above mentioned defect. This algorithm allowed much faster detection speeds at some costs of accuracy compared to Faster R-CNN, Single-Shot MultiBox Detector (SSD) and Retina-Net detection models which we also tested.

[0054] Further, there is the surveillance to detect wrong or non-homogeneous temperature distribution of the metal sheet surface in a contact free manner during production. For detecting wrong or non-homogeneous temperature distribution of the metal sheet surface, a spectral imaging device, i.e. IR thermographic camera, is used to monitor correct and homogeneous temperature of metal sheet. For detecting this a simple linear regression model with two output layers is used, wherein the first output layer represents the absolute temperature and the second output layer represents the temperature homogeneity from 0, i.e. not homogeneous, to 1 , i.e. fully homogeneous / uniform. OK- data were generated during production with parallel supervision by a hand held mobile thermal camera. Non-OK-data were recorded on a sample metal sheet that was treated with ice cubes and blow torch to create temperature inhomogeneities.

[0055] In a step 2, an adhesion promoter is applied. The adhesion promoter being a reactive (1 K or 2K) liquid is applied on one or both inner sides of metal cover sheet, wherein many different application methods exist for applying the promoter like a moving nozzle, a spray nozzle, a drip beam, a rotating washer disc, etc. In this step, there is the surveillance to detect fully or partially missing of the adhesion promoter as well as non-uniform distribution on the metal sheet(s). For detecting this, optical cameras are used to cover full width of metal sheet that has been covered with adhesion promoter. Video / image data from multiple cameras are stitched together to result in full image of full width of metal sheet. This detection is provided by a machine-learning, ML, network with Resnet (Residual neural Network) architecture configured to detect deviation from ideal adhesion promoter coverage of metal sheet. The model was trained with OK-data from normal production runs. Within these runs adhesion promoter distribution was supervised and controlled by human visual inspection and the amount of adhesion promoter was controlled by flow measuring followed by off-line gravimetric control. Defect data, i.e. nonOK data, were recorded during regular start up and stopping procedures of the process. This automatically generates very non-uniform and defect rich videos and images. In a step 3, liquid raw material(s) for insulation foam core are applied. The foam raw materials are pre-processed and pre-mixed before application. The reactive mixture is applied on full width of the lower metal sheet, wherein many different application methods with different outlet geometries exist like nozzle, multiple nozzles, multiple nozzles on a bar (“poker”), pre foaming pipes, etc.. The application device can be fixed or moving from side to side of the metal sheet, i.e. oscillating movement. All application methods aim on covering the full width of the metal sheet with the reactive raw material. In this step, the surveillance is to detect a (premature, partial, full) blocking of the raw material outlet device resulting in defects or non-uniform ity of the raw material distribution which then in turns may lead to defects in the finally produced foam core. For this detection, in case of fixed outlet device, an optical camera is used to cover full width of device and full size of leaving raw material flow until hitting the lower metal sheet. In case of a moving output device, the camera can be either mounted fix either to cover full width of the oscillating area or more preferable mounted in a way that it moves together with the output device. For the detection, an approach with two machine-learning, ML, models are used. In a first step, it is used a very shallow and fast Mobilenet R-CNN clone to detect the actual position of the output device (region of interest ROI). In a second step, an Autoencoder type model (Encoder + Decoder) is used and the reconstructed image with the input image by weighted sum of squared pixel differences are reconstructed. A (premature, partial, full) blocking of the outlet is detected, when a difference is above a threshold. Both networks are trained on OK-images during human supervised normal production runs. Defect images are not needed with this approach.

[0056] In a step 4, raw materials are formed to a final foam core. The liquid raw materials start leveling out and foaming up (chemical or physical foaming). The foam expands until it is confined by hitting to top metal sheet (side areas are confined by moving temporary blocks). At the same time chemical reaction starts cross-linking and hardening of the foam structure. In this step, there is the surveillance to detect inhomogeneities in foam leveling and defects in the growing foam, e.g. bubbles, blow offs, ripples, etc.. For the detection, video / image data are provided by optical cameras to cover full width of raw materials starting to foam on the lower metal sheet. The Detection is provided by a Fast classification model, e.g. VisionTransformer is perfectly suited for this job. OK data is generated during normal operation. Defect data with respect to bubbles and blow offs are collected during process starting up. Defect data with respect to foam leveling was created by intentional using false (higher) temperatures of metal sheet and raw materials. In this step, there is the surveillance to monitor foaming speed, i.e. foam expansion, gel time, i.e. chemical reaction speed, and contact time and contact position, i.e. foam hitting the upper panel. For the monitoring is provided by optical measurement devices like laser distance meters, laser profilers (2D, 3D) or 3D industrial cameras might work. Due to confined space and ATEX zone constraints, we use a 3D laser profiler. The monitoring is provided by a LSTM (Long Short Term Memory) model to analyze the time series of foaming speed, gel time, contact time and contact position. Make prediction of future development of those values based on actual and historic (time series) data. Training of the network was done during normal operation each time the operator changed production parameter and kept them constant afterwards for more than 1 day of production. Training data was collected during more than 4 weeks of production time. With this approach, we get training start measure data results for nearly all relevant production parameters.

[0057] In a step 5, there is provided a pressing and curing of sandwich during foam hardening.

[0058] In a step 6, there is provided a cutting / sawing of endless sandwich into panel size.

[0059] In a step 7, a cooling down of storage of panels is provided. Panels with reacted foam core are stored until they are cooled down to room temperature. Before, during and after this step the panels move very slowly, which allows to find an ideal position to check final panel quality. In this step, there is the surveillance to detect dimensions and shape, (buckling, bulge, convex or concave bending). For this, two 2D laser profilometer are used at the entrance of the cooling down area. One profilometer from top side, one profilometer from bottom side of the panel, wherein the panel is moving through / between profilometers letting both flat sides being scanned by profilometers. Full 3D shape data can be extracted by combining top 2D profile (measured from above) with lower 2D profile (measured from below). Combining can be done for example by subtracting the profiles with given, calibrated distance between top and bottom 2D profilometer. As the panel moves continuously between the profilometers the 2D data recorded over time can be transformed to 3D by calculating the missing length parameter z in moving direction by z = velocity ■ time. Especially the absolute thickness and the homogeneity in thickness as well as any bending can be easily tracked from the 3D shape data. The detection is provided by a simple sum square of deviations algorithm to compare the measured 3D shape (2D profiles over time) with the given ideal geometric shapes. Simple threshold gives feedback of panel being in spec or out of spec.

[0060] In a step 8, a final confectioning, milling, cleaning of panel is provided.

[0061] In a step 9, a packaging and storing / preparing for transport.

[0062] Fig. 4 illustrates a box diagram of a system for monitoring a complex process having at least two process steps. The system 30 for monitoring a complex process having at least two process steps, comprises a first providing unit 31 for providing monitoring data by at least one monitoring process for each one of the at least two process steps. The first providing unit 31 includes a machine learning algorithm, a neuronal network, a measurement, a Region based Convolutional Neural Network, an Al model, a machine learning network with a residual neuronal network, a machine learning model, a classification model and / or a sum square of deviations algorithm. The system 30 further comprises a second providing unit 32 for providing aggregation data. The second providing unit 32 includes an aggregation layer, wherein the aggregation data are provided by the aggregation layer by performing summarizations and / or combinations based on all of the provided monitoring data, in particular the aggregation data are further provided by a comparison of the provided monitoring data with reference data. The aggregation layer is a machine-learning algorithm. The system 30 further comprises a third providing unit 33 for providing instructive data based on the aggregation data. The instructive data includes alarms, interruptions of the complex process, errors in the complex process and / or recommendations how to react on interruptions, alarms and / or errors.

[0063] Optionally, the system 30 further comprises a fourth providing unit 34 for providing intermediate aggregation data by an intermediate aggregation layer based on a subset of the provided monitoring data. The fourth providing unit 42 includes an intermediate aggregation layer. The providing of intermediate aggregation data is provided by the aggregation layer by a machine-learning algorithm. Optionally, the system 30 further comprises a fifth providing unit 35 for providing control data for controlling or adapting the complex process and / or at least one process step of the complex process.

[0064] The present disclosure has been described in conjunction with a preferred embodiment as 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 a different nodes using different equipment / data processing units.

[0065] In the claims as well as in the description the word “comprising” does not exclude other elements or steps and the indefinite article “a” or “a” 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.

Claims

Claims1 . A computer-implemented method for monitoring a complex process having at least two process steps, comprising: providing monitoring data by at least one monitoring process for each one of the at least two process steps; providing aggregation data by an aggregation layer by performing summarizations and / or combinations based on all of the provided monitoring data; providing instructive data based on the aggregation data.

2. Computer-implemented method according to claim 1 , wherein the aggregation data are provided by the aggregation layer by a comparison of the provided monitoring data with reference data.

3. Computer-implemented method according to any one of the preceding claims, wherein the monitoring data of each one of the at least two process steps are provided by a separate monitoring process of the at least one monitoring process.

4. Computer-implemented method according to claims 1 or 2, wherein the monitoring data of at least two process steps of the at least two process steps are provided by a same monitoring process of the at least one monitoring process.

5. Computer-implemented method according to any one of the preceding claims, wherein providing of aggregation data by the aggregation layer is provided by an algorithm, in particular a machine learning algorithm.

6. Computer-implemented method according to any one of the preceding claims, further comprising: providing intermediate aggregation data by an intermediate aggregation layer based on a subset of the provided monitoring data; wherein providing of aggregation data by the aggregation layer is based on the provided intermediate aggregation data.

7. Computer-implemented method according to claim 6,wherein providing of intermediate aggregation data by the intermediate aggregation layer is provided by an algorithm, in particular a machine learning algorithm.

8. Computer-implemented method according to any one of the preceding claims, wherein the monitoring process is at least one out of a group, the group consisting of a machine learning algorithm, a neuronal network, a measurement, a Region based Convolutional Neural Network, an Al model, a machine learning network with a residual neuronal network, a machine learning model, a classification model and a sum square of deviations algorithm.

9. Computer-implemented method according to any one of the preceding claims, wherein, when the at least one monitoring process is a plurality of monitoring processes, the monitoring processes are different to each other.

10. Computer-implemented method according to any one of the preceding claims, wherein the instructive data includes at least one out of a group, the group consisting of alarms, interruptions of the complex process, errors notifications in the complex process and recommendations how to react on interruptions, alarms and errors, respectively.

11. Computer-implemented method according to any one of the preceding claims, further comprising: providing control data for controlling or adapting at least one process step of the complex process based on the instructive data.

12. Computer-implemented method according to any one of the preceding claims, wherein the complex process is at least one out of a group, the group consisting of a chemical process, a continuous process, a manufacturing process, a band-reactor process, a manufacturing process of sandwich panels, a manufacturing process of foams, a manufacturing process of duroplastic foams, a manufacturing process of thermoplastic foams, a manufacturing process of polyurethane, a manufacturing process of polyisocyanorate, a manufacturing process of polystyrene, and a manufacturing process of melamine resin.

13. An apparatus for monitoring a complex process having at least two process steps, the apparatus comprising: one or more computing nodes; and one or more computer- readable media having thereon computer-executable instructions that are structured such that, when executed by the one or more computing nodes, cause the apparatus to perform the following steps: providing monitoring data by at least one monitoring process for each one of the at least two process steps; providing aggregation data by an aggregation layer by performing summarizations and / or combinations based on all of the provided monitoring data; providing instructive data based on the aggregation data.

14. Use of control data generated by the computer-implemented method according to claim 11 for controlling or adapting the complex process and / or at least one process step of the complex process.

15. Computer program element with instructions, which, when executed on computing devices of a computing environment, is configured to carry out the steps of the method according to any one of the claims 1 to 12 in an apparatus according to claim 13.

Images