DE202021004135U1 - Location-based component analysis based on a digital component twin - Google Patents

Abstract

A computer program for processing data of a manufacturing process, comprising instructions which, when executed by one or more processors, cause the one or more processors to perform operations comprising:
Acquisition (510) of process data from one or more steps of the manufacturing process for a component that is used in the one or more steps of the manufacturing process, such that the process data can be assigned to a point on the component at which the component is in a certain manufacturing step is processed, the location being specified as a 3D spatial coordinate;
acquiring (520) location data indicative of a location on the component with which the process data is associated;
storing (540), in a database, the process data together with the location data to obtain a data set of a digital twin of the component, the process data and the location data being structured via a location index; and
Using the digital twin data set to analyze (550) the manufacturing process.

Description

field of invention

The invention relates to the acquisition, storage and analysis of data from a manufacturing process, among other things for the purpose of visualizing digital component twins with analysis options. The technologies described here are primarily used in the field of discrete manufacturing, for example in the manufacture of aircraft wings or car doors.

In accordance with the provisions of the Utility Model Act, only devices as defined in the attached claims for protection, but no processes, are protected and the object of the utility model. Insofar as reference is made to methods below in the description, these references only serve to explain the device or devices protected in the attached claims by way of example.

background

The implementation of the "digital factory" is still a challenge for manufacturing companies. Available solutions for data collection and analysis have their origins in the IT sector and are therefore geared towards the use cases known from the IT sector, for example whether individual servers in a server farm have error messages or to detect server failures, etc. Recently, exactly these solutions are also used to monitor machines in factories, e.g. to detect error messages and failures. This results in productions being monitored in the same way as servers, which has major disadvantages.

State-of-the-art IT solutions use a timestamp to collect and structure production data. As a result, only time-based analyzes can be carried out with regard to a machine. This enables the following queries, for example: How productive was the machine? How much downtime was recorded? When and how often were there error messages? Etc.

With known tools for collecting and processing production data, analyzes with regard to a component or component type are impossible or can only be carried out with an extremely large amount of resources. This applies, for example, to queries such as: Why is every fifth component defective at a certain point? Does the machine speed have an impact on the quality of my component? Etc.

Due to this limitation, productions still analyze and optimize their part quality manually by experienced process experts. This means that production is extremely dependent on individual employees who know their machines well and intuitively improve the process. This unsystematic approach creates significant additional costs due to inefficient processes, poor part quality and the risk that certain employees are unavailable.

In other words, the limitations with regard to component analyzes are based on the fact that sensor values and machine parameters that are generated during the production of components are not collected, stored and structured in the prior art in such a way that questions about component quality can be answered quickly and easily be able. Production data is typically only saved with a time stamp. If you now want to get to the bottom of a phenomenon at a specific location in the component, e.g. to investigate why a pore or delamination occurs at a specific location on the component, you usually need the production data from the component location where the phenomenon occurs occurs. A time stamp would only help indirectly if the machine position is stored somewhere with a high-precision, synchronized time stamp, which in most cases is not even available, and the various sensor values and machine parameters can be combined via the time stamp. With this analysis strategy, however, extremely high computing power would be required for large amounts of data, so that this procedure is not practicable for large amounts of data with conventional computers. One reason for this difficulty is that time series databases used in conventional systems for analyzing production data are indexed and sorted by time. A search for a position in order to find the associated time stamp and thus to find the process parameters sought would be highly inefficient. This makes it impossible, for example, with high-frequency process measurement values using known means to compare two identical components in relation to their location, i.e., for example, a question of the type "Having component 1 at the point (x, y, z) in component 1 at the time of production the same process conditions prevail as at the same point in component 2 at the time of the production of component 2?”.

The cause of difficulties in analyzes with regard to individual components or a component type is that either location information from production data is not available from today's IT solutions is recorded, stored and used for analysis and thus location-based analyzes with regard to a component cannot be carried out or would generate considerable additional effort in data processing, or the location information is not used to structure the data in this way and make it searchable. The term location information refers here, as will be explained below, to the point at which a component is processed in a specific production step.

The structuring of location data is used, for example, in the field of geographic information. Geo data is stored, for example, as a point cloud ("point cloud"), for example for the purpose of visualizing a 3D scan of a landscape or using longitude and latitude (i.e. 2D) for analysis. Technologies such as the indexing of 2D location data in a key value store are used for analyses. However, such management structuring of production data, especially not with 3D information, is not known in the prior art.

In the prior art, for example in connection with 3D printing methods, it is known to record production data at X-Y positions in certain layers during the processing of a workpiece. This is not full 3D data in that direct comparison with data from other 3D sources is difficult when other 3D data sources, such as computed tomography (CT) data, represent full 3D locations and associated data that are also between individual layers can lie. In this context, a data comparison in 2 dimensions separately for each layer is common, but this does not represent an original comparison in a 3D data room. Layer-based data acquisition and analysis is only common for certain manufacturing processes and is therefore limited in its application.

A comparison of actual and target values relating to the dimensions of a workpiece or other properties at specific points on the workpiece is known from conventional methods for processing and analyzing production data. To determine target values, reference is made to a reference model and/or CAD model of the workpiece, for example. In the novel methods described herein for structuring and processing production data, data from different production steps of a workpiece are compared based on their location information, with the respective production data and location data being recorded during the respective process steps. A CAD model is not absolutely necessary here. Likewise, the methods described herein are not limited to the comparison of actual and target values (the latter taken from a reference or CAD model).

It is thus an object of the present invention to provide methods and means for systematically analyzing and improving the quality of components.

summary

The invention is defined in the independent claims. Dependent claims specify embodiments of this invention.

One embodiment relates to a computer-implemented method for processing data from a manufacturing process, comprising: acquiring process data from one or more steps of the manufacturing process, for a component that is used in the one or more steps of the manufacturing process, such that the process data of a location can be assigned to the component on which the component is processed in a specific production step; capturing location data indicating a location on the component to which the process data is associated; storing, in a database, the process data together with the location data to obtain a data set of a digital twin of the component, the process data and the location data being structured via a location index; and using the digital twin dataset to analyze the manufacturing process.

In series production, the assignment of process data to specific locations of a component enables, for example, the location-specific comparison of production data for components, preferably of the same type, the detection of deviations and errors in components, and the analysis of error causes and the determination of machine maintenance.

In one embodiment, the acquisition of the process data and the acquisition of the location data are triggered substantially simultaneously or in response to a master trigger event.

This has the advantage that, as a response to the master trigger event, data read out simultaneously or essentially simultaneously can be assigned directly to a specific location coordinate on the currently manufactured component, and thus a fusion of production data, i.e. a merging of data that is can be assigned to essentially the same location of a manufactured component belong together, but are recorded, read out and stored independently of sensors, can be simplified.

In one embodiment, a time of acquisition is stored in the database with the process data and the location data.

This has the advantage that production data that is recorded at the same time or essentially at the same time is recognized later in a merging step as belonging together using an essentially the same or the same time stamp.

In one embodiment, the process data includes measured values whose spatial assignment corresponds to the location of a measurement on the component. For example, when milling, the rotational speed of the milling spindle would be recorded at a specific location and defects in a later visual inspection could be traced back to fluctuations in the rotational speed.

In one embodiment, collecting process data further includes: receiving measurement data from one or more sensors monitoring the manufacturing process and/or parameters of one or more machines used in the manufacturing process; wherein the measurement data include the measured value and the time of a measurement and can each be assigned to a location on the component at which the measurement data was recorded; and wherein the parameters include the value and time of use of the parameters and can each be assigned to a location on the component at which the machine was used.

This has the advantage that the conditions prevailing at the time of production and the set machine parameters determined by the measured values can be traced in detail for each individual component. This is particularly helpful for the quality and error analysis of individual components and for determining the causes of errors, since, among other things, it can be determined what influence machine parameters have on measured values.

In one embodiment, location data is the basis for the location index, the database being a spatial or spatio-temporal database, i.e. a database whose implementation is designed to efficiently store, search and analyze spatial or spatio-temporal data, and optionally an implementation of the location index based on an N-tree.

The location index enables efficient access to related production data, i.e. production data that influenced the production result at a specific location of a component in the course of a manufacturing process. In other words, the method described here and the structuring of the process and location data described here using a location index enable a quasi “backward search”, i.e. the search for properties during a production step at a specific location of a component and not just, as in conventional ones Process, the place or point at which a certain property of a component or a state occurred during production ("forward search"). This possibility of "backward search" starting from a certain place to the properties of this place at certain production steps opens up the possibility of completely new applications, which are described in the following.

A "forward search", which is possible in the methods described here as well as in conventional methods for structuring and processing of production data, aims, for example, to determine for a specific value in the production data at which point of a component and in which manufacturing step this is value, for example an excessive temperature or too low a speed during processing. For certain applications, this "forward search" is not expedient for the following two considerations: First, many components are produced with different settings. For example, the speed of a robot making a component by milling is fast in one place and slow in another. It is therefore not expedient to search in the production data for a robot speed that has fallen below a certain level, since it may be desirable to fall below a certain speed at certain points on the workpiece. In other words, it can be envisaged to program a robot in such a way that it works at low speed at some points on a workpiece and at high speed at other points. Thus, a method based on a "forward search" for detecting supposed defect indications (e.g. slow robot speed) would be far too tolerant (if a significantly too low speed were assumed to be critical, defects could be overlooked), or too imprecise (if if too high a speed were assumed to be critical, many non-defective points could be identified as possibly defective). A location-selective method for comparing and searching for production data is therefore desirable for such an application. Second, it is often not known which properties in production data are advantageous on the one hand and indicative of defects on the other. Knowing what values and qualities are considered beneficial or indicative of one Defects to be evaluated can only be determined when properties from other, for example downstream, process steps are known and are correlated with process data from the previous manufacturing process. For example, data from a subsequent investigation using computed tomography can determine possible defects and their positions or location on the workpiece or component and determine possible causes of the defect in the production process through correlation with location-based production data and properties. This new method can only be carried out if process data from different process steps can be efficiently compared "over the place" or point by point with precise location.

The use of a location index in the present application area of production data acquisition and analysis based on data structures that are used, for example, in the area of geo-databases, is illustrated further below: When using a conventional table and index structure, e.g. a table with production data comprising one column for time and one column each for x, y and z coordinates, it is possible to specify specific entries for a given x, y and z position. However, without a location index, you have to search the entire table once to determine which x,y,z value is closest to the position you are looking for. In typical cases, a value with identical coordinates will not appear in real data. This would require a large number of search steps to be carried out, of an order of magnitude that is linear in the number of rows in the table. For example, if you want to compare all items from a first table with all items from a second table and each of these tables has 1 million entries, it would mean that you would have to perform 1 million to the power of 1 million (= 1 trillion) search steps. To do this, you have to apply the Pythagorean theorem to each comparison to calculate the distances for each line to the point you are looking for, which includes, for example, a root operation ("square root"), 2x squaring and an addition. In typical cases with large datasets, it could mean that the calculations on conventional computers would take several months, if not years, which is why conventional methods of organizing and storing production data are not suitable for, for example, efficiently comparing production data in a 3D coordinate field .

In one embodiment, the method also includes: merging of process data, which can each be assigned to the same location, based on detected location data; and/or merging of process data that was acquired at the same or essentially the same point in time based on a point in time of the acquisition of the respective process data.

Fusing process data enables the use of key-value databases as an efficient implementation for storing and retrieving manufacturing data for the analyzes described herein.

In one embodiment, a measured value indicates a temperature and/or density of the component at the respective location of the measurement. For example, measurements from a computer tomographic examination can be compared with measured values from the manufacturing process of a component, for example to quantify pore formation due to temperature influences.

In one embodiment, the process data is temperature data and the parameters include power consumption, speed, rotation and/or contact pressure of one or more machines used in the manufacturing process. In this way, the connection between influencing parameters can be examined and quantified in order to make monitoring less sensitive and to produce less superfluous rejects, since, for example, in processes such as the carbon fiber deposit, temperature, speed and contact pressure are closely related (with higher speed required higher contact pressure and temperature are required, with lower contact pressure the temperature must not be too high, etc.). Thus, a significantly more efficient analysis is possible than simply monitoring threshold values for individual parameters.

In one embodiment, the power consumption for a laser machine is measured at a specific frequency at which a metal part of the component melts, and the data on the power consumption of the laser machine is further mapped to spatial coordinates that indicate where the melting of the metal part on the component takes place takes place at a certain point in time. Among other things, this makes it possible to check whether defects are related to fluctuations in the laser power.

In one embodiment, the manufacturing process is an additive, subtractive, or transforming process and the one or more machines include at least one mechanically moved machine.

Additive, subtractive or forming processes on a component require high mechanical precision when guiding the processing work stuff. Mechanically moved machines are subject to wear and require constant monitoring and adjustment in order to ensure the consistently high quality of the manufactured components. Precisely in this area, the precise quality control of individual components after production is an essential factor for the early and precise detection and determination of the need for machine maintenance. Since the condition of the machine can change in relation to its location, eg due to its own weight and lever, determining differences in machine behavior only makes sense at the same or similar locations. For example, it can be normal for a robot to run with a power consumption between 3.9 kVA and 4.05 kVA over the entire workspace. It is then not possible to estimate whether parts are worn, since the range of fluctuation is considered normal over the entire processing of a component. At a dedicated location of a component type, the range of power consumption for different components of the component type is typically much smaller, so that deviations among the individual components can be identified and thus maintenance needs of a machine can be predicted much earlier and more precisely.

In one embodiment, the analysis of the manufacturing process includes one or more of the following: partitioning of a space to be visualized into discrete spatial areas, for each spatial area pre-evaluation of production data assigned to the respective spatial area, the pre-evaluation being based on information from the location index and the location index being aggregated information about measurement data contains; Comparative analysis of the digital twin with a digital twin of another component of the same component type; and comparing the data set of the digital twin of the component of a component type with a reference model for the component type; and visualization of the digital twin or optionally visualization of individual spatial areas of the digital twin according to the results of the preliminary evaluation. As an option, aggregated information is recorded according to the pre-evaluation in the location index and provides information, for example, about a sum of measured values, the arithmetic mean of the measured values, the minimum or maximum measured value in a (partial) spatial area of the spatial or spatio-temporal production data.

In one embodiment, the reference model is created based on a plurality of data sets of digital twins of components of the component type in such a way that the reference model specifies statistical values of corresponding process data from the plurality of data sets of digital twins. Common features of components, normal values of production data and permissible deviations are thus recorded in the reference model. When comparing a specific component or comparing the production data recorded in its virtual component twins with the reference model, this enables atypical deviations beyond the standard values recorded in the reference model to be identified and identified as possible causes of quality defects in the specific component.

In one embodiment, the reference model is the result of one or more simulations that make a prediction for specific locations on a component. This allows simulated components to be compared in detail with real components in order to determine deviations in both directions and use them for optimization. In one direction, it can indicate that the component is faulty. In the other direction, it can indicate that the simulation does not yet depict reality accurately enough.

In one embodiment, the comparative analysis is used to determine deviations in manufacturing accuracy among multiple components of the same component type, and results of the comparative analysis are used to predict machine maintenance and/or to determine a quality of a component. As a result, maintenance work on the machine or repairs to the component can be planned better and the probability of quality shortfalls in the final component due to statistically distributed errors in subsequent production steps can be calculated in order to preventively repair or reject components at an early stage.

One embodiment relates to a device for processing data from a manufacturing process, comprising: data processing device; and storage device in communication with the computing device, wherein the storage device stores instructions that, when executed on the computing device, cause the computing device to perform operations comprising: acquiring process data, from one or more steps of the manufacturing process, for a Component used in the one or more steps of the manufacturing process, such that the process data can be assigned to a location on the component; capturing location data indicating a location on the component to which the process data is associated; storing, in a database, the process data along with the location data to obtain a digital twin record of the component; and using the digital twin dataset to analyze the manufacturing process. Other embodiments of the device further include instructions that the device device according to the computer-implemented method for processing data of a manufacturing process described herein.

One embodiment relates to a computer-readable storage medium comprising instructions which, when executed by a computer, cause the computer to carry out the following steps: Acquiring process data from one or more steps of the manufacturing process for a component used in the one or more steps of the manufacturing process is such that the process data can be assigned to a location on the component; capturing location data indicating a location on the component to which the process data is associated; storing, in a database, the process data along with the location data to obtain a digital twin record of the component; and using the digital twin dataset to analyze the manufacturing process. Further embodiments of the computer-readable storage medium further comprise instructions that cause the computer to take further steps according to the computer-implemented method described herein for processing data of a manufacturing process.

character list

• 1A shows a component with a marking of a specific location on the component.
• 1B shows a graph indicating the temperature of the component during the manufacturing process along the time course of manufacture.
• 1C shows the visualization of a 3D model of a component, ie a digital component twin, in which each location of the component is color-coded according to a temperature measured there at a specific point in time (e.g. the point in time at which the laser melts the powder in metallic 3D printing).
• 2 shows a view of a component twin with detailed representation of data from the manufacturing process associated with a specific location of the component.
• 3A-E show the step-by-step acquisition of data from a manufacturing process for a component based on the visualization of a corresponding component twin, including data that has already been acquired.
• 4 shows a flow chart of a method for acquiring data of a manufacturing process.
• 5 shows a flow chart of a method for acquiring and processing data from a manufacturing process.
• 6A and 6B each show a visualization of different component twins of the same real component.
• 6C shows a visualization of an overlay of the digital component twins 6A and 6B .
• 7A shows a real component with an error.
• 7B Figure 1 shows a computer and application for rendering part digital twins of the real part 7A for error analysis.
• 8th shows a flow chart of a method for processing data of a manufacturing process.
• 9 shows a flowchart of a method for creating a reference model.
• 10 shows a flow chart of a method for the statistical analysis of a reference model.
• 11 Figure 12 shows a flow chart of a method for constantly revising or optimizing a reference model.
• 12 shows an overview of a system for manufacturing a component, including a data processing device and storage device for processing data from the manufacturing process of the component.

Detailed description

In order to provide systematic analyzes and improvements with regard to the quality of components, an essential aspect of the technology described herein, together with production data on a component, is also a location on the component and/or a time of processing or processing a component at a respective Location of the component to capture. The collection of this data enables so-called location-based analyzes of individual components, of series of identical components, and the machine function and wear. In other words, the structuring of production data according to a component position makes this data usable for the creation of digital component twins for applications and analyzes in the production environment. This collection of digital component twins and their application to perform different analyzes of a manufacturing process is described in more detail below.

A manufacturing process may be, for example, an additive, subtractive, or transforming process and, for example, may involve one or more machines, at least one machine of which is a mechanically moved machine.

The term data source refers to devices, such as sensors, that generate data that can be directly or indirectly assigned to a specific location in a component. These can be measured values from external sensors, from internal sensors, target values of the system control, machine settings (e.g. a speed specification), etc.

The terms component and workpiece are used here essentially synonymously. The term component position denotes a location in the component, point in the component, location in the component, areas in the component, position (x, y, z).

The term production data is essentially used synonymously with the term data of a manufacturing process and/or with the term process data. Production data include, for example, measured values whose spatial assignment corresponds to a measurement at the respective location on the component, for example a temperature, a pressure, and/or a density of a component at the respective location of the measurement. Measurement data can also be, for example, observed machine speed, power consumption of a laser. Also included are parameters of one or more machines that are used in the manufacturing process, e.g. control data of a machine, i.e. set machine speed, contact pressure, etc.

For location-based analysis, production data of a component are structured and stored based on location, in order to enable evaluations of specific locations in the component in real time, ie in the order of a few milliseconds to a few seconds. The motivation for location-based data and the principle of location-based data and its visualization is the example of 1A until 1C illustrated.

1A 12 shows a physical component 110 with a marking of a specific location 120 on the component. Possible production data that can be associated with this point of this specific individual component are, for example, one or more temperature measurement values at one or correspondingly more production steps of this component.

1B FIG. 13 shows a graph 130 which indicates a temperature of the component 110, for example over the entire production step, along the time course t of the production of the component. At the representation in 1B temperature and time can be determined, but not the precise location on the component where the temperature was recorded.

1C shows the visualization of a 3D model of a component, ie a digital component twin, in which each location of the component is marked visually, ie with a different color or shade of gray, according to a temperature measured there at a certain point in time. At the representation in 1C location and temperature can be determined as well as the time. However, it is not possible to search for data from a specific point in time the other way round.

The inventive location-based acquisition and storage of any production data for any in the form of "digital component twins" enables the visualization of these component twins as 3D point clouds, and associated analyses.

The structuring according to location data means that a multitude of process data for a single component is stored and managed in such a way that all process data, i.e. process data from different production steps, can be efficiently retrieved for a specific location on a specific individual component.

Technologies for storing and managing this data are, for example, so-called key-value database systems (“key-value database”) such as RocksDB and Couchbase, or generally No-SQL databases. Specifically, a unique value is calculated from the spatial coordinates and optionally a unique identifier of a component, which is used as a key in a data record of the key-value database. Optionally, the unique is a hash value. That key (“key”) is associated with a value (“value”) that specifies associated production data, for example measured values and/or machine parameters. Individual measured values and/or machine parameters are optionally in turn associated with corresponding time stamps at which the values were recorded or set. A key-value database system can contain production data from a single component, from multiple components of one component type, or from multiple components of different types. In the latter two cases, the data organization must be such that data associated with a single component can be efficiently identified and selected. For example, a component's ID may be encoded into the key.

As an alternative or in addition to a key-value database, other database systems can be used for storage and administration as long as they enable the following essential properties. It is essential for such systems and the underlying data model that data can be retrieved efficiently for a specific location on a specific component. In a possible implementation, data is organized as a key-value pair, and a series of data is made efficiently searchable via a corresponding index. This index is referred to herein as a location index, since the indexing takes place via the key, which essentially reflects location information.

This structuring of production data according to their assigned location information makes it possible to quickly and easily visualize and search for process data in a production environment with regard to their location in the component. A possible representation of the data belonging to a single component is a kind of 3D point cloud, which is referred to as a “digital component twin”.

Storing process data together with the location data within a database thus includes structuring the process data and the location data using a location index.

2 shows a view of a component twin with detailed representation of data from the manufacturing process associated with a specific location of the component. With location-based data structuring, for a specific location in a component (e.g. upper right corner) of the component in 2 Process parameters such as the milling spindle speed or the temperature can be output or these, for example, in a three-dimensional false color representation, as in 5 shown, visualize. The figure shows a specific point, the spatial coordinates (y, x, z), as well as the width and height of a spatial area (width, height), and statistical information on process parameters and measured values such as temperature, machine speed, contact pressure. In this example, a time stamp is also recorded as well as any marking (tag). For a clear visualization, the space is partitioned into discrete sections along each of the three dimensions and thus, for example, divided into cuboid or cube-shaped spatial areas. For each of these spatial areas, pre-evaluations of the production data contained therein, ie production data assigned to a location within the cuboid, are formed, eg statistical evaluations, mean or min-max values of all data contained in this spatial area. The visualization, ie the coloring of the spatial area, is then defined according to the values determined in the pre-evaluation. This has the advantage that, depending on the size or granularity of the partitioning into spatial areas, the resolution of the 3D visualization can be adapted to computer performance, display options or user requirements.

3A-E show the step-by-step acquisition of data from a manufacturing process for a component based on the visualization of a corresponding component twin, including data that has already been acquired. The position coordinates of a tool are, for example, retrieved from a machine during the manufacturing process of a component and fused with sensor or machine data (e.g. the temperature of the component or parameters for controlling the tool). In 3A initially only a few process data are recorded, in 3B until 3E a more comprehensive set of process data is visualized in each case, which corresponds to the progress of the manufacturing process of the specific component at successive points in time.

In one embodiment, the process data collection and the location data collection occur substantially simultaneously and are triggered in response to a master trigger event. The use of a master trigger means that in addition to sensors that record sensor values, there is also a source for location data (e.g. x, y, z). The master trigger thus triggers data acquisition at at least two data sources, for example at one sensor (source 1) and another source (source 2) which acquires the location at which the sensor (source 1) recorded the sensor data. Both are triggered by the same electrical signal from the so-called master trigger. The second source (source 2) can be, for example, a manufacturing machine that typically “knows” or constantly records its own position or that of the end effector. Alternatively, the second source (source 2) could also be a device such as a laser tracker (light time-of-flight measurement), or a technology based on a triangulation measurement method, etc.

A time of acquisition can also be stored in the database together with process data and location data. This means, for example, that each source records its data with a time stamp, especially the source that records the location. On the basis of this time information, known as a time stamp, production data can then be merged, i.e. data that is recorded from different sources essentially at the same time can be recognized as belonging together and sensor values or machine values can therefore be assigned to location information recorded at the same time. This so-called merging of process data, which was recorded at essentially the same point in time, is based on a point in time when the respective process data was recorded.

When recording production data, the location reference can be created directly or indirectly: For example, in the case of thermal camera recordings of a component, the location of a measured value can be determined via the pixel coordinates. The location reference, ie the association of sensor data with location data, can thus be produced directly. More generally, this means that a machine already records and transmits its production data together with location information. This is conceivable, for example, with industrial robots. Process data that can each be assigned to the same location can be merged based on recorded location data.

The technologies described here can be used primarily in discrete component manufacturing or the inspection of components, assemblies or finished products. For example, there are systems that record the temperatures layer by layer with a certain resolution during the production of metallic 3D printed components and save them as an image. The technologies described herein can use these images as input to map measurement data to location based on the image data and thus create a digital component twin where we store the temperatures for each location in the component (see for example 2 and 3 ).

However, as explained above, the location reference can also be produced indirectly via the use and comparison of time stamps by assigning production data with the same or finitely the same time stamps to a location recorded at this point in time.

The fusion of data via a common location, i.e. the association of data that is recorded at a specific location on a single component in the course of a production process, can be computationally intensive, for example in the case of indirect association via a time stamp, and the processing and selection among a large amount of production data. The merging therefore does not usually take place in a real-time request, but is a step in connection with the storage and processing of production data, which takes place nearer to or also at a later time than the acquisition of the production data.

The process for the step-by-step acquisition of production data when manufacturing a component in one or more steps is shown in the flow chart in 4 shown.

In step 410, a manufacturing process starts for a component A whose data are to be recorded. Step 420 represents a manufacturing step, i.e. the machining or processing of the component itself. For each step 420 of the manufacturing process, sensor and machine measurement values and parameters are recorded simultaneously in step 430 and stored together with a reference that allows later calculation of the location at which component allows, or directly with a component coordinate. If a reference is detected, this reference is converted to a respective component coordinate in a subsequent step 440 . A reference can be a timestamp, for example. Location coordinates, which are also recorded with a time stamp, then allow the indirect assignment of production data and measured values to location coordinates via the reference to a time stamp that is or is approximately the same time. Step 440 is followed by an optional transformation 450 of the data into a coordinate system of the respectively processed component A. The transformation is required, for example, if different coordinate systems are used for two different data sources. For example, if Source 1 data is location information from the robot and Source 2 data is coordinates from a laser tracker. Then you have to transform the coordinates of one of the two systems into the other (e.g. via a homogeneous transformation matrix). The transformation thus makes it possible to efficiently compare and visualize production data from the same components (different data sources, e.g. temperature and density), which are recorded with spatial coordinates in different coordinate systems. Furthermore, it is possible to correlate data from different components (with the same data sets).

Instead of storing the recorded production data or their values on the respective coordinates, the values could already be projected onto a surface during input/transmission. This eliminates the possibly computationally expensive coordinate transformation in the data analysis. However, this step is only relevant in special manufacturing processes and limits the analysis due to the loss of the actual coordinates.

This parallel execution of a production step 420, recording of production data and determination 430 and conversion 440 and 450 of assigned component coordinates is repeated 460 for one or more steps in the production of component A, until finally the production of component A is completed 470.

The method according to FIG. 4 records all process data, including measured values, machine parameters, etc., from the same and/or different manufacturing process steps for each individual component of a specific type and a specific location, for example specified by a spatial coordinate (x, y, z), component coordinate, assigned in the component.

5 shows the complete process of collecting, storing and analyzing process data.

In step 510, process data from one or more steps of the manufacturing process are recorded for a component in such a way that the process data can be assigned to a location on the component. Assignable means that the location data is recorded directly with the individual process data (example of the thermal camera recording). Alternatively, an assignment can be made using a reference, such as a time stamp.

The gathering of process data in step 510 includes, for example, receiving measurement data from one or more sensors that monitor the manufacturing process. Measurement data here include, for example, the measured value and time of a measurement, and can each be assigned to a location on the component at which the measurement data was recorded, the parameters including the value and time of use of the parameters and can each be assigned to a location on the component which the machines were used. The acquisition of process data can further include the acquisition of parameters of one or more machines used in the manufacturing process, for example control data of one or more machines used in the manufacturing process, such as set speed, power consumption, a rotation and/or a contact pressure . For example, the power consumption for a laser machine can be measured at a certain frequency at which a metal part of the component melts. The data on the power consumption of the laser machine is also mapped, directly or indirectly as previously described, to spatial coordinates that indicate where the melting of the metal part on the component is taking place at a given point in time.

In step 520, location data is collected. In the example of a thermal camera recording, steps 510 and 520 coincide. Steps 510 and 520 correspond to the repeated application of individual steps 430 to 450 4 .

Position coordinates can be determined as follows: Instead of real coordinates (i.e. from a measurement), the position coordinates could also be interpolated from a machine program (e.g. NC program) for controlling a movable machine. Instead of retrieving the coordinates via the production machine, external coordinate measuring devices (e.g. laser trackers, cameras, etc.) could also be used.

In step 530, process data that can each be assigned to the same location are merged. Fusion means determining process data that can be assigned to the same component location in order to save this determined data in a subsequent step in such a way that later efficient access to this data in real time is possible via a location index. The merging can take place directly, based on location data recorded with the production data, or indirectly via a reference determined location information, e.g. when process data recorded at essentially the same point in time are merged. The merging can take place immediately when the data is recorded, or at a later point in time after the data is recorded.

In step 540, process data is stored in a database along with the location data to obtain a record of a digital twin of the component. For example, the merged data from step 530 is stored in a key value store or database storage engine. In this case, clustering according to location coordinates and spatial sorting, eg sorting according to spatial coordinates, can be carried out in order to create a location index. A request for production data relating to a specific component and a specific location can be found efficiently using the location index with just a few search steps. This makes it possible with manageable effort to create data sets, as in 2 and 6 shown to visualize, or as in 7 shown to compare different components of the same type (e.g. temperature values of two identical components or temperature and robot speed values of one component) with each other. The visualization for the purpose of the analysis is possible, for example, in 3D as false colors and also possible with a superimposed visualization of a CAD model of the component to improve the orientation and assignment of the production data to the component.

In step 550, collected production and location data is used for analysis purposes.

• • Die erfassten Fertigungsdaten (Stromverbrauch des Lasers) als Falschfarben 3D-Bild anzeigen lassen, um z.B. nach Hot-Spots zu suchen (siehe zum Beispiel 2 und 3). Wobei Schnitte durch das Bauteil in jeder Orientierung möglich sind.
• • Die erfassten Fertigungsdaten zu filtern, um z.B. nur Daten über oder unter einem bestimmten Schwellwert zu visualisieren.
• • Temperaturdaten und Daten des Laser-Stromverbrauchs korrelieren, um zu analysieren ob es einen Zusammenhang gibt.
• • Wenn das Bauteil später im Betrieb geprüft wird (z.B. durch CT oder Ultraschall) kann ein Experte die Daten aus der Fertigung des konkreten Bauteils verwenden und mit den Prüfdaten vergleichen um z.B. festzustellen ob bestimmte Prozess- oder Messerwerte eine Indikation für spätere Bauteilmängel wie z.B. Risse sein können.

In a production scenario, for example, the power consumption of a laser tool used in production and a temperature of a surface of a component are measured at a specific frequency and combined with the location at which the laser is currently melting the material. With the technologies described here, the following analyzes are possible, for example:

• • The recorded production data (power consumption of the laser) as a false color 3D image displayed, e.g. to search for hot spots (see for example 2 and 3 ). Whereby cuts through the component are possible in any orientation.
• • Filter the recorded production data, for example to only visualize data above or below a certain threshold.
• • Correlate temperature data and laser power consumption data to analyze if there is a relationship.
• • If the component is later tested in operation (e.g. by CT or ultrasound), an expert can use the data from the production of the specific component and compare it with the test data, for example to determine whether certain process or measurement values are an indication of subsequent component defects such as cracks could be.

More generally, possible analyzes in step 550 are described in detail below.

A first possible analysis concerns the visualization of one or more digital twins of a specific physical component. This analysis will be in 6A until 6C illustrated.

For each individual physical component, all process data is saved with a location marker or location index (x, y, z) in each production step, so that all available data sets for a component can later be superimposed and compared, as described below.

For example, component twin 610 indicates temperature data of a particular first machine in the production process. For example, in a metallic 3D printing process, a temperature sensor for a specific component provides a temperature value for the melt pool (the point at which the laser just melts the metal) in a grid or with a resolution of 0.125mm x 0.125mm detected. Based on the current machine position, these dimensions can be transformed into real coordinates, which can then be transferred to a general coordinate system, eg the drawing coordinate system, ie the coordinate system of the visualization. This results in a digital temperature component twin 610 based on a first data set with temperatures for each location in the component. Different temperatures are shown in the illustration in 6A shown in different shades of gray.

In a second process step, eg computed tomography, location-based data is again generated for the same physical component. This data is transformed into the same coordinate system as the data from the other recorded process steps. Thus we get a second (or further) digital component twin 640, shown in 6B , based on a second dataset for the same physical component.

Since the different component twins are each indexed using the same location coordinates, each value of the data set can be superimposed and compared with each value of the second data set, as in 6C shown, or if the data sets do not have exactly the same position, the closest data points are compared with each other, or different statistical values for certain spatial regions.

7A shows a real, physical, component 710. Assuming a certain point of the component with coordinates (x, y, z), a defect is detected after the production of the component.

7B Figure 7 shows a computer and application 720 for visualizing digital component twins. In this representation, 730, 740, 750 are different component twins of one and the same real component 710 7A .

For example, component twin 730 indicates temperature data of a particular first machine in the production process. The component twin 740 indicates, for example, the speed setting of a second machine involved in the manufacture of the component during the manufacturing process of the component 710 . The component twin 750 indicates, for example, a rotational angular velocity of a robot axis of the second machine during the production of the component 710. From the visualization of the component twins and the process data recorded with it, it can then be recognized, for example, that the values 735 and 745 of the component twins at the points (x, y, z) are normal, but the value 755 at the same point of the component twin 750 deviates from the usual values. Thus, according to the underlying production data of the component twin 750, the one deviation in the angular velocity of rotation can be determined as the probable cause for the manufacturing defect on the physical component 710. Of course, the cause of an error can also be determined from the combination and/or correlation of production data from different digital component twins of one and the same physical component. For example, an interesting question could be whether the quotient of the set laser power and the measured heat is the same at these points in the component.

This comparability is also possible if different process data cannot be assigned to exactly the same location. For example can n mean values can be calculated over certain areas of space (represented as cubes). Furthermore, very efficient nearest-neighbor searches are possible thanks to the location-based structuring and storage of the data. In other words: On the basis of this location-based data structuring, different types of data analysis can be carried out, which result in a variety of applications and benefits in manufacturing. This has the great advantage that data with slightly different coordinates can also be compared efficiently. This is absolutely necessary for most manufacturing processes, since machines cannot repeat their movements exactly down to every decimal place. Machine movements always deviate from each other, even when manufacturing the same components.

Offline analyses

In the following, some examples of analyzes are presented, which are essentially based on the comparison of digital twins and are carried out independently of the production of a component, i.e. offline after the completion of the production of a single component or a series of components of the same type. Offline analysis can compare twin parts (datasets) that come from different data sources or sensors and may have been collected at different steps in the manufacturing process.

8th shows a flow chart of a method for processing data of a manufacturing process. Specifically, the method processes data associated with specific locations of a component to generate an off-line statistical analysis of the cause of a non-conformance or failure. The process in 8th begins at step 810, where at a particular point (x,y,z) on component A there is a defect that is to be investigated, or the cause of which is to be investigated. In step 820, all data (eg, temperature, velocity, pressure values) from all process steps at point (x,y,z) of part A is retrieved. In step 830, the data from a particular data source (eg, temperature) within component A is compared. For example, for each data type, a comparison is made to determine whether the value at point (x,y,z) differs significantly from corresponding data at other points on component A. For example, one can analyze whether the temperature values at a point to be examined in component A are abnormal compared to temperature values at other points in component A. The aim is to find a parameter (in this case the temperature) that is the site to be examined behaves abnormally. For example, the temperature readings of the component could be anywhere between 595-601°C, only at the point to be examined at 605°C. Or the component could be manufactured at an average temperature of 599°C and 95% of all values could be within a range of less than +-1 °C, only at the point to be examined would the value be outside - albeit still within acceptable range, but it would be statistically significant. In step 840, data from different data sources are compared to the same data sources from another component (of the same type A). For example, if an error occurs at location (x, y, z) in part ID2 (of part type A), all data at this location (or a range around this location) can be retrieved and with all the same data from parts of type A to be compared. In this way, it is possible to check whether, for example, several components at location (x, y, z) were manufactured with the same temperature as component ID2 or whether the temperature of component ID2 at location (x, y, z) was statistically abnormal. The analyzes do not always have to relate to the same location. In step 850 it is also possible to use data from similar situations for comparison. For example, in milling, one could check whether the temperature of a tool was responsible for a certain damage by comparing data from a similar situation. For example, a parameter radius could be specified and stored at each point. If you later want to investigate why a milling cutter has damaged a certain point, in addition to comparing with components of the same type at this point, the points for comparison at which the radius value is the same could also be used to analyze whether the milling cutter also caused damage at these points. For example, if a machine has just milled a 10 cm radius and an error has occurred (eg the material is becoming brittle), the data from this point could be compared with the data from other identical or similar radii with the current data. In step 860, a statistical analysis of the cause of the deviation or error is made.

Instead of the comparison with target specifications (e.g. from the NC program that controls the machine when manufacturing the component), the comparison values could also come from a simulation. As explained below, a simulation enables or supports the creation of a reference model of a component.

Embodiments of the invention may include other aspects. For example, component twins (data sets) generated from different data sources can also be correlated to create an offline correlation of data sets. In the above example, the data from the same data source is compared and examined. In the correlation analysis data from different data sources, from one or more steps of the manufacturing process, at certain locations or at all locations are compared with each other. This makes it possible to determine a correlation of different parameters within a component within a process. For example, it is known that when pre-impregnated carbon fibers are deposited, the temperature, speed and contact pressure have an influence on the placement or properties of the component. The correlation analysis can be used, for example, to check which constellation of parameters (e.g. for speed, temperature and contact pressure) leads to good or bad quality by additionally entering or analyzing the points at which bad quality has occurred. This could also lead to a location-related result, for example that a different combination of parameters has to be used in the area of an incline in the component than at the edge of a component or in the middle of a component.

Another aspect concerns the correlation of different parameters within a component over several processes. Correlation analyzes can be very effective for cross-process component investigations. For example, a temperature component twin from a metallic 3D printing process can be correlated with the results of a computed tomography (CT) examination without making a specific assumption. For example, it could be investigated whether there is a connection between any process parameters (e.g. the temperature) and the density differences (from the CT). In addition, the target-actual value pairs could also be examined, where the target values equal the parameter settings and the actual values equal the parameter measurements). For example, it could be analyzed whether the set speed leads to the same actual speed at all positions in the component or whether the connection is always the same or whether this is abnormal at a point in the component that is conspicuous.

The aforementioned off-line analysis aspects of the invention offer a number of advantages and technical effects. For example, systematic error cause tracing is enabled. It is well known that currently practiced production methods and processes only have a limited understanding of the causes of defects and only allow limited possibilities for their analysis. This leads to sometimes high and constant quality costs. For example, companies report up to 30% scrap in the production of small series from metallic 3D printing, partly due to the flat learning curve. The embodiments of the inventive concept described here enable quality defects in a component to be traced back to their origin and rectified quickly. This saves productions considerable repair and scrap costs.

The structure of process data as a digital component twin makes a targeted search for the causes of errors and quality deviations very easy. For example, if an operator wants to examine a point in the component more closely because he has noticed an error or a sensor has directly measured the quality of the component, he can specifically analyze the manufacturing history of the component at these points. In this way, if the operator has detected pores at location (x,y,z) in the CT examination, he can retrieve all data from all production steps that have been stored, correlated or interpolated for this location or an area.

A further technical advantage of the present invention is that the determination of influencing factors in the components is made possible. Especially when new process steps are introduced, productions find it difficult to specify the relationships between different process parameters or to determine the influencing variables of the individual parameters. The structure and processing of data from a production process as a digital component twin makes it very easy to search for adjustments that have an impact on the quality of the component.

Furthermore, the present invention enables error tracing throughout the entire product life cycle. Offline analyzes are possible both in production and in the later life cycle of the component when the component is already in operation (e.g. installed in an airplane or a vehicle). If it is recognized during operation that components have a defect, and this leads to a recall (e.g. in the automotive industry) or grounding (e.g. in the aviation industry), the present invention could be used to quickly find out which components have the same production parameters were manufactured, and specifically to recall the corresponding vehicles. This would also enable continuous long-term documentation of components across multiple production companies (e.g. across suppliers and OEMs).

Reference model and online analysis

In addition to being used for offline analysis, the location-based method of processing data can also be used for an "online analysis" to create a reference model for a to create a certain type of component and to use it in a targeted manner. This section is about assessing the quality of components and the stability of processes online using a master model or reference model. For this purpose, the data from at least two components are used and statistical values are calculated for each point or area of the components (e.g. mean value, standard deviation, maximum value, minimum value, critical value, etc.) and a new component reference model is created from this, Also referred to as the reference model in the following. This reference model no longer contains the individual temperature values for each individual component, but instead, for example, the average value of the temperatures measured for each point in the room. This is referred to below 9 until 11 explained. In a component reference model, reference is made to data that relates to specific positions on a component. In contrast to the component reference model, a component-independent reference model, which is to be distinguished from the component reference model, is also described in the further course of the detailed description.

A reference model is created in order to be able to compare many components with one another or to check online during production whether the current component is manufactured in the same way as previous components with sufficient quality. In doing so, not simple measured values are stored at the coordinates, but statistical values such as mean value and standard deviation. When a new component is to be tested (possibly even during its manufacture) it is easy to check what its deviation is compared to the standard deviation of the reference model at the respective location.

9 FIG. 12 shows a flow chart illustrating a process for creating a reference model of the manufacturing process of a component during one or more steps of the manufacturing process. In step 910, each area of each component is evaluated, eg by an expert or automatically by a quality measurement. If a certain amount of components has been produced or their data has been included in the reference model, only the points that are outside the tolerance are evaluated in step 920 . Here, the process data of a first component is classified according to the quality of the data. If a reference model is to be created from the data of the very first component, a user can also directly specify parameters for calculating statistical values, e.g. it could be specified that each temperature, i.e. for each recorded point in space, may vary by 5% around a certain value . However, if there is not enough data for the reference model, a second component of the same type as the first component is created in step 930 and the corresponding process data is saved. In step 940, the process data from the second component are classified according to their quality. In step 950, the process data of all components of the same type that have already been produced are used to calculate the reference model with statistical values for each location of a component of the same type. This statistical approach becomes very complex when there are interactions between factors that have an impact on the quality of the component. In such cases, machine learning methods based on the data processing of the manufacturing process could learn relationships and thus calculate tolerance values for certain points in the component, possibly based on a combination of different data, e.g. if a certain temperature leads to damage in the component, but only if a certain pressure is applied to the component at the same time, etc. In addition, the reference model can also be based on calculated variables. For example, in carbon composite production, it is often not of great importance what temperature a heat source of a depositing machine has, but much more what temperature input the component experiences. This depends not only on the temperature of the heating source, but also on the dwell time and speed of the machine. In some cases, it is not the direct data that is saved, but derived variables such as the temperature input.

In summary, a reference model is created based on a plurality of data sets from digital twins of components of a specific component type. The process data recorded in the reference model correspond to statistical values of corresponding process data of digital twins.

The reference model of a component can also be the result of one or more simulations in order to define the location-related tolerance fields for specific locations on a component. The creation or refinement (the latter if in combination with the previously presented method for creating a reference model) of the reference model via a simulation has the advantage that the precision of the reference model can be created much faster and before the actual production begins. In addition, the simulation can then be refined the other way round using the reference model and a better prediction can be made with regard to the necessary tolerance fields for the next component.

10 is a flowchart depicting a process for statistically analyzing a manufacturing process of a component during one or more rerer steps of the manufacturing process illustrated, wherein the analysis according to an embodiment of the present invention on the reference model created by 9 based. The generated reference model can be used to compare data of a currently produced component online during production. Regarding 10 , in step 1010 the user will determine a reaction if a new component deviates from the reference model. In step 1020, another third component of the same type as the first component is produced and the corresponding production data is stored. In step 1030, the production data of the third component are compared with the reference model and, if necessary, the predetermined reaction is triggered. For example, if in a 3D printing process for a new component the point (x, y, z) was melted with a temperature of 504.21 °C, the reference model can be used to check whether this value is between a min and max value of the reference model is at this point or by how much sigma this deviates from a mean value. If the value falls outside a tolerance window of the reference model, the process can be stopped, an alarm given, or the location noted. In step 1040, the process branches in the event that the reference model is to be further modified, for example. A further modification of the reference model is necessary, for example, if the desired statistical accuracy of the model is not yet sufficient with the data processed so far. If further components are to be manufactured, the process in step 1050 iteratively branches to step 1020. Otherwise, production ends in step 1060.

11 shows a flowchart of a method for constant revision or optimization of a reference model 9 described. The user defines a metric or tolerance window with which currently produced components are monitored, eg whether the data of the production process and the parameters are in a 2-sigma range of the reference model. The result is production data that is within the required interval and production data that is outside of this interval. In step 1110 it is decided either by the user or an algorithm (eg, based on quality measurements) whether the component has a sufficient quality even with the position outside the tolerance window or not. The components, or just points that have led to good quality despite violation of the tolerance window, can then improve the reference model (step 1120), for example by adjusting the mean value or the maxima of the reference model. The components or the points that have violated the tolerance and where bad quality was also found can also be transferred to the reference model, for example in order to define or strengthen the limits of the tolerance. At step 1130, the analysis process continues iteratively at step 1020 if there are more components to analyze or fabricate. Otherwise, in step 1140, the manufacturing process ends.

Embodiments of the invention may include other aspects. For example, the reference model can be used to introduce new machines. With reference models of existing machines, production can set up a new machine from the point of view of the component to be manufactured. The reference models of a part, created on existing machines, can be used as benchmarks for the new machine and the new machine settings and parameters can be defined using the reference part. On an existing machine, all process measurement values that appear significant for the production of a component (e.g. energy input) are recorded, saved and used later, regardless of the specific machine settings (e.g. it could be that for a machine 1 a temperature was set and the power of the heating source was set for a machine 2). The digital reference model is based only on the process measurements. When a new machine is run in, the user can call up the process measurement values for a new component and compare them with the reference model. The employee can then iteratively adjust the settings of the new machine with the aim of "allowing the same process measurement values to "arrive" on the component as in the reference model. Furthermore, all existing data is recorded. The running-in of a new machine could be automated or at least supported during running-in. For this purpose, the relevant process measurement values for a component for the new machine and the parameters of the machine that led to these process measurement values are recorded. Machine learning algorithms can then be used to learn how the parameters have to be modified in order to achieve the desired result (i.e. the same process measurement values as on the component on the run-in system).

Another aspect of the invention is the use of reference models for flexible production changes. The reference model offers the user the opportunity to quickly and easily observe the influence of changed process parameters and to optimize them with just a few iterations. The technical approach here provides that relevant data that is to be observed is marked in the reference model. This offers the Users have the opportunity to optimize the process with regard to the other parameters. For example, it could be that the user would like to increase the machine speed for a carbon fiber placement, since it is generally known that the temperature input into the component should remain constant. Therefore, the temperature of the heating source must also be increased. The present invention can be used to check whether after the process modifications the relevant quantities or derived or calculated quantities remain constant. If the speed is not marked as critical, changing it will not result in a message. If the temperature input is marked as critical at the same time, a message will appear because the temperature input is lower at higher speeds because the machine dwell time is shorter with the same heating output. An increase in the heating temperature with an increase in speed would again lead to production without a message if the ratio was correct.

The on-line analysis aspects of the invention mentioned above offer a number of advantages and technical effects. For example, the online comparison of the reference model and the component during production makes it possible to monitor component quality during production and, if necessary, to stop production early and avoid expensive further processing of defective components. The advantage over conventional sensors, which monitor a specific parameter in a specific process, comes into play when several parameters (in one or more subsequent process steps) are dependent and an incorrect relationship between these parameters can cause component errors.

Another technical advantage of the invention is the reduction of machine lead times. The introduction of new machines is very expensive, even if it is the same type of existing machines. The machine settings can never be adopted one-to-one. The time required to set the new machine correctly manually is therefore considerable.

The present invention works even if the existing and the new machine differ greatly, since only the process measurement values (ie the result) of the component to be produced are used for the reference model. It is generally reported that large production machines (e.g. riveting systems, fiber laying systems, etc.) require up to 1.5 years to run in, even for well-known, experienced machines. With the present invention, customers can introduce new machines with few iterations and monitor changes in near real-time and improve them if necessary. This leads to flexible production changes.

12 . Shows an overview of a system for manufacturing a component including a data processing device and storage device for processing data of the manufacturing process of the component. The component 1210 is processed by a moving machine 1220 . In the example shown, a data processing device 1240 records machine parameters and measured values from a sensor 1230. The sensor 1230 can record, for example, location data of the tool guided through the machine 1220. Device 1240 may also control moveable manufacturing machine 1220, for example. Recorded production data are communicated via a network 1250 to a data processing device 1270 and stored there in a key-value database, which is stored, for example, in the storage device 1280 connected to the data processing device. 1260 Displays a terminal for visualization analysis of manufacturing data as described herein. The terminal is communicatively coupled, for example via network 1250, to data processing device 1270 and storage device 1280. Data processing device 1240 and 1270 may be, for example, a conventional computer with appropriate peripherals and network connectivity, and may be implemented on separate or shared hardware. Network 1250 may be wireless or wired, and may be implemented using conventional network technologies, for example. Alternative system architectures based on virtualized machines, networks and storage systems, as well as implementations based on a cloud computing infrastructure are within the grasp of the expert.

Component-independent reference data and online analysis

All mechanically moving machines (e.g. milling machines, robots, riveting systems, etc.) suffer from positional inaccuracies that increase over time. Productions therefore maintain their machines at regular intervals in order to avoid breakdowns and correct position inaccuracies. In recent years, attempts have been made to reduce this maintenance to a necessary minimum with approaches to preventive maintenance. In general, these approaches focus on irregularities in the machines and are structured as follows: in a time series database, for example, the power consumption of robot axes is stored. This measurement gives the power consumption over time. If the robot now always the same trajectories in always the same As time goes by, it is possible to find out whether the power consumption for a new component has changed or “deteriorated” by comparing the power consumption curves. Of course, this only works if it is always the same type of component and the process always runs at the same speed. Statistical values can also be formed for a process step in order to determine deviations over time. However, in order for a deviation to be recognized, this presupposes that the power consumption has changed significantly. A disadvantage of these approaches is that only symptoms of the machine are monitored, but not the result on the component itself. In addition, a machine behaves very differently, e.g. robot axes in different poses of the robot can have very different power consumption even with a well-maintained robot. If all axes of the robot are stretched out, the power consumption changes simply because of the greater leverage of the weight of the axes. In summary, the comparative analysis is used to determine deviations in manufacturing accuracy among a plurality of components of the same type of component, and results of this comparative analysis are used to predict machine maintenance and/or to determine a quality of a component.

Due to the location-based data processing method, the present invention offers other options with which the disadvantages mentioned above can be ignored. For example, the inventive concept takes into account, but is not limited to, the component, the component quality, and the associated process measurements. For this purpose, the digital twins that may have been created for different components from different component types in the "processing room" are used and superimposed. This results in reference data for different positions in the entire movement space of the system, not necessarily related to a component, but to the entire work space - hence "component-independent reference data", with which, for example, the power consumption at certain points in the room can be monitored.

With the approach of the present invention, this results in the advantage that conclusions can be drawn about the machine even during the production of various components. It is virtually irrelevant whether component type A or B is manufactured, since the power consumption of the robot axes at a position (x, y, z) is only checked as long as the position (x, y, z) has already been measured . With each additional component, the machining area could be “explored” further. In the best case, by a certain point in time there will be enough data to define a reference value for each point in space (with a certain resolution), e.g. how high the power consumption at this point should be. The method could also be used for completely new components (and thus deduce the need for maintenance based on process values). The approach of the present invention could also be further refined by considering not only power consumption but also speed and then possibly calculating a factor per point in space. For example, if the robot of component type A processes e.g. at a higher speed than component type B. This of course changes the power consumption of the robot axes, which is recorded in the digital "space twin".

This novel creation of spatial data according to the present invention has the following advantages for a manufacturing company: With preventive maintenance ("predictive maintenance") based on component quality, any machine can be monitored along the entire production chain with regard to its position accuracy by analyzing the quality of the components and not the machines themselves. A machine is only serviced when the quality of the components produced decreases. This allows regular maintenance intervals to be replaced by maintenance as needed. Such on-demand maintenance is also more efficient than traditional approaches to preventive maintenance for machines.

Implementation of a location index

The implementation of a location index is based, for example, on an N-tree, i.e. a higher-dimensional extension of a binary tree. In the special case of a spatio-temporal database, it is a three- or four-dimensional index structure; with the usual three spatial dimensions and one temporal dimension. Special forms of these trees are known, for example, as R trees, quad trees or oct trees. The tree nodes are created in a special way and stored in a part of the binary key-value database, which way of creation also links them to the tree structure.

The coordinates of individual spatial values in the database are converted to a different form for the calculation of their binary key in the database. Each individual coordinate is, for example, first converted to a signed binary fixed-point number with a known bit length, for example 32 or 64. This is done by first converting the global coordinates into a local coordinate system by means of a reversible transformation in order to to optimally use the applied value range of the selected fixed point representation.

In this local coordinate system, the coordinates are "cut off" to form a local coordinate key in fixed-point representation. Finally, the local coordinate key is converted into an index on a multidimensional space-filling curve, a z-curve. To do this, the bits of the various fixed point coordinates are rearranged in an interlaced manner, so that one or more bits follow each other for each coordinate. The binary representation of this point index, starting with the most significant bits, is used as part of the key in the database.

Based on this, the data index is defined as follows: Each entry in the index tree structure corresponds to an axis-parallel space cube (voxel) in local coordinates and contains additional aggregated information about all data points contained therein. The space cube, or cube for short, is denoted by a prefix of the binary index number and includes all coordinates beginning with that particular prefix. The binary key in the database consists of the concatenation of the prefix length, i.e. the number of bits contained per coordinate, and the prefix itself. The prefix length is equivalent to the depth of the cube in the tree structure. This allows easy reconstruction of the keys of the parent node and the child nodes, i.e. the tree structure, and allows access to these nodes, if any, in the database. The root is uniquely denoted with a prefix length of zero and an empty prefix.

The described voxel can also be read in local coordinates from the binary key. In addition, by reversing the local transformation, the global volume of the voxel can be described as a polytope in global coordinates for each entry. It is also recorded how many data points are spatially inside the cube. For indexed scalar data, a so-called cache cube also contains the sum of all data points, the arithmetic mean of all data points, a lower and an upper limit for their values and the minimum and maximum coordinates that actually occur. In other words, aggregated information in the location index provides information about the sum of measured values, the arithmetic mean of measured values, the minimum or the maximum measured value in a sub-area of spatial or spatio-temporal production data. A pre-evaluation of information based on the location index can thus take place by means of the cache cubes, so that the index includes aggregated information, in the form of the cache cubes, about measurement data. Furthermore, each cube stores basic information about its child nodes. This includes in particular which child nodes are present.

When searching the tree to execute a query, all the information in a cube is used to decide which part to search on. In particular, it is taken into account whether there are any data points in this part at all and whether loading a child node is expedient. In addition, it is evaluated whether descending to a child node is more efficient than directly iterating over the contained data points and whether this iteration should take place in a concurrent routine.

Using the cached data of a cache cube, logic predicates specified in the database query can also be checked in some cases for all data points in the cache cube at once. In special cases, for example, it can be guaranteed that the result is 'wrong' for all points contained, without evaluating the predicate individually for all points contained. Then the cube including all its children can be skipped in connection with a search for points where the predicate returns 'true'. The avoided loads save memory bandwidth. In addition, the reduced number of evaluations reduces the required processor time.

For some cache cubes, because of the nature of their parent nodes or their child nodes, or the data points they contain, it can be calculated in advance that the query evaluation algorithm will never choose to load those cache cubes. These cubes can still be deleted from the database when the index is created. This saves memory that would otherwise be required for representing the aggregated values. For example, it is not beneficial for a browsing algorithm to load a cache cube with a single data point, since that cube contains no information beyond the data point.

Other implementations

In example implementations of the technology described herein, a computing device 1240, 1270 may include any number of hardware and/or software applications. In example implementations, one or more I/O interfaces may enable communication between the computing device and one or more I/O devices. For example, a universal serial bus port, a serial port, a floppy disk drive, a CD-ROM drive, and/or the one or more user interface devices, such as a display, keyboard, keypad, mouse, control panel 1260, touch screen display, microphone, etc., that enable user interaction with the computing device. The one or more I/O interfaces can be used to receive or collect data and/or user instructions from a variety of input devices. The received data may be processed by one or more computer processors, as desired in various implementations of the disclosed technology, and/or stored in one or more storage devices 1280.

One or more network interfaces may facilitate connection of the inputs and outputs of the computing device to one or more suitable networks 1250 and/or connections; e.g. B. the connections that enable communication with any number of sensors 1230 associated with the system 1200. The one or more network interfaces may further facilitate connection to one or more suitable networks 1250; for example, a local area network, a wide area network, the Internet, a cellular network, a radio frequency network, a Bluetooth enabled network, a Wi-Fi enabled network, a satellite network, any wired network, any wireless network, etc. for communication with external devices 1220 and/or systems.

Numerous specific details are set forth in the present specification. However, it is understood that embodiments of the disclosed technology may be practiced without these specific details. In other instances, well-known methods, structures, and techniques have not been shown in detail so as not to obscure the understanding of this specification. References to "one embodiment", "example embodiment", "some embodiments", "particular embodiments", "various embodiments", etc., indicate that the embodiment(s) of the disclosed technology so described have a particular feature, a may include a specific structure or characteristic, but not every embodiment necessarily includes the specific feature, structure, or characteristic. Furthermore, repeated use of the phrase "in one embodiment" is not necessarily referring to the same embodiment, although it may be.

In describing example embodiments, certain terms are used for clarity. Each term is intended to have its broadest meaning as understood by those skilled in the art and to include all technical equivalents that operate in a similar manner to achieve a similar purpose.

While particular embodiments have been described herein, the present technology is not limited to the example embodiments disclosed, but encompasses various modifications and equivalent arrangements. Although specific terms are employed herein, they are used in a generic and descriptive sense only and not for purposes of limitation.

Examples are used in this written description to disclose specific embodiments of the disclosed technology, including the best mode, and to enable any person skilled in the art to practice the disclosed technology, including making and using any suitable devices or systems and performing any integrated methods . The scope of the claims below may include other examples that are within the purview of a person skilled in the art, even if they include equivalent structural elements with insubstantial differences from the literal language of the claims.

Claims (15)

A computer program for processing data of a manufacturing process, comprising instructions which, when executed by one or more processors, cause the one or more processors to perform operations comprising: Acquisition (510) of process data from one or more steps of the manufacturing process for a component that is used in the one or more steps of the manufacturing process, such that the process data can be assigned to a point on the component at which the component is in a certain manufacturing step is processed, the location being specified as a 3D spatial coordinate; acquiring (520) location data indicative of a location on the component with which the process data is associated; storing (540), in a database, the process data together with the location data to obtain a data set of a digital twin of the component, the process data and the location data being structured via a location index; and Using the digital twin data set to analyze (550) the manufacturing process. computer program claim 1 , wherein the acquisition of the process data and the acquisition of the location data simultaneously in response to a master trigger event to be fired; and/or wherein a time of acquisition is stored in the database with the process data and the location data. Computer program according to one of the preceding claims, wherein the process data include measured values whose spatial assignment corresponds to a location of a measurement on the component. Computer program according to one of the preceding claims, wherein the acquisition of process data further comprises: receiving measurement data from one or more sensors monitoring the manufacturing process and/or parameters of one or more machines used in the manufacturing process; wherein the measurement data include the measured value and the time of a measurement and can each be assigned to a location on the component at which the measurement data was recorded; and wherein the parameters include the value and time of use of the parameters and can each be assigned to a location on the component at which the one or more machines were used. Computer program according to one of the preceding claims, where location data is the basis for the location index, the database is a spatial or spatiotemporal database. computer program claim 5 , wherein the operations further comprise: merging (530) of process data, each of which can be assigned to the same location, based on detected location data; and/or merging (530) process data that was recorded at the same point in time based on a point in time of the recording of the respective process data. Computer program according to one of claims 3 until 6 , wherein a measured value indicates a temperature and/or density of the component at the respective location of the measurement. Computer program according to one of Claims 4 until 7 , wherein the process data are temperature data, and wherein the parameters include a power consumption, a speed, a rotation and/or a contact pressure of the one or more machines used in the manufacturing process. computer program claim 8 wherein the power consumption for a laser machine is measured at a specific frequency at which a metal part of the component melts, and further wherein the data on the power consumption of the laser machine is mapped to spatial coordinates which indicate where the melting of the metal part on the component takes place takes place at a certain point in time. computer program claim 8 or 9 , wherein the manufacturing process is an additive, subtractive or forming process, and/or wherein the one or more machines comprises at least one mechanically moved machine. Computer program according to one of the preceding claims, wherein the analysis of the manufacturing process comprises one or more of the following: Partitioning of a space to be visualized into discrete spatial areas, for each spatial area pre-evaluation of production data assigned to the respective spatial area, the pre-evaluation being based on information from the location index and the location index containing aggregated information about measurement data; Comparative analysis of the digital twin with a digital twin of another component of the same component type; and Comparing the data set of the digital twin of the component of a component type with a reference model for the component type; Visualization of the digital twin, optional visualization of individual room areas of the digital twin according to the results of the preliminary evaluation. computer program claim 11 , wherein the reference model is created based on a plurality of data sets of digital twins of components of the component type and data recorded in the reference model correspond to statistical values of corresponding process data of digital twins; and/or wherein the reference model is the result of one or more simulations that make a prediction for specific locations on a component. computer program claim 11 , wherein the comparative analysis is used to determine deviations in manufacturing accuracy among a plurality of components of the same type of component, and results of the comparative analysis are used to predict machine maintenance and/or to determine a quality of a component. Device for processing data of a manufacturing process of a component (1210), comprising: data processing device (1270); and A memory device (1280) in communication with the computing device, wherein the memory device stores instructions that, when executed on the computing device, cause the computing device to perform operations comprising: acquiring process data from one or more steps of the manufacturing process, for the component that is used in the one or more steps of the manufacturing process, such that the process data can be assigned to a point on the component at which the component is processed in a specific manufacturing step; acquiring location data indicating a location on the component to which the process data is associated; storing, in a database, the process data together with the location data to obtain a data set of a digital twin of the component, the process data and the location data being structured via a location index; and using the digital twin dataset to analyze the manufacturing process. A computer-readable storage medium comprising instructions which, when executed by a computer, cause it to: Capturing process data from one or more steps of the manufacturing process for a component that is used in the one or more steps of the manufacturing process in such a way that the process data can be assigned to a point on the component at which the component is processed in a specific manufacturing step becomes; acquiring location data indicating a location on the component to which the process data is associated; storing, in a database, the process data together with the location data to obtain a data set of a digital twin of the component, the process data and the location data being structured via a location index; and Using the digital twin data set to analyze the manufacturing process.

Images