Method for producing a product

Abstract

The invention relates to a method for producing a product, said method having the steps of: providing specification data of the product, said specification data containing reference values for product features of the product; processing the product in a plurality of processing steps in accordance with the specification data; creating a test program which comprises a plurality of test criteria, each of which is assigned at least one of the product features; testing the product in accordance with the test criteria of the test program by testing the product features assigned to the test criteria, each reference value assigned to the product features being compared with corresponding actual values actually determined during the test, and collecting test data which indicates the result of the test and / or the actual values of the product features.

Description

[0001] Method for manufacturing a product

[0002] Description

[0003] The present invention relates to a method for manufacturing a product according to product specification data, wherein the method includes a step of testing the product based on test criteria of a test program. The invention further relates to a computer program for carrying out such a method.

[0004] In industrial manufacturing, quality control is a crucial process that ensures the consistency, reliability, and safety of manufactured products. Regular inspection of products or intermediate products is an integral part of the manufacturing process, aiming to minimize defects and maximize quality. This process extends across various stages of production, with final inspection being a critical phase.

[0005] In an automotive plant, for example, quality control plays a particularly important role, as vehicles are complex products comprising a multitude of components and systems. Final inspection on an automotive production line marks the end of the manufacturing process and involves a thorough examination of each vehicle to ensure it meets stringent quality and safety standards. This includes checking various aspects such as mechanical functions, electronic systems, safety features, aesthetic standards, and other criteria. Quality assurance in automotive manufacturing is crucial for identifying defects before vehicles leave the plant, thus ensuring high customer satisfaction and compliance with legal regulations. This inspection can be carried out manually by an inspector or automatically or semi-automatically.

[0006] For more complex products, such as automobiles, it is not possible to compare all product characteristics with the respective specification data during every quality inspection, especially during final inspection at the end of production. Instead, effective quality assurance is achieved by creating a test program that includes a selection of test criteria for verifying a range of product characteristics, but not all of them. Experienced quality assurance personnel define and select the test criteria for the test program and / or base it on statistics and empirical data regarding frequently occurring defects or failure-prone components. Particularly safety-critical components, legal regulations, and other constraints are also taken into account.

[0007] Nevertheless, testing, especially for complex products, remains largely random, and predicting defects and tolerance deviations with satisfactory accuracy is not yet possible in practice. This is due not only to the complexity of the product characteristics but also to interactions between components, component defects, and the occurrence of specific events during the manufacturing process. For example, if a component identified as defective is replaced or reworked during a processing step, this replacement or rework can affect the probability of defects in other components of the product. For instance, the processing step of installing a seat in a vehicle body carries a known risk of damage to the vehicle's interior.If, during the manufacturing process of a vehicle, a seat needs to be replaced because a defect is detected in the seat during an intermediate testing step, the probability of damage to the vehicle's interior due to the additional seat installation increases. Such correlations are difficult to account for, even for experienced inspectors.

[0008] It is therefore an object of the present invention to provide a method and a computer program for manufacturing a product which also enables improved quality control for complex products.

[0009] According to a first aspect of the invention, this problem is solved by a method for manufacturing a product, the method comprising: (a) providing specification data of the product which contains reference values ​​for product characteristics of the product, (b) processing the product in a plurality of processing steps according to the specification data, (c) creating a test program which comprises a plurality of test criteria, each of which is assigned at least one of the product characteristics, (d) testing the product according to the test criteria of the test program by testing the product characteristics assigned to the test criteria, comparing the reference values ​​assigned to each product characteristic with the corresponding actual values ​​of the product characteristics actually determined during the test, and (e) recording test data which indicate a result of the test and / or the actual values ​​of the product characteristics.in the step of creating the test program, the test criteria of the test program are determined using a KL system.

[0010] According to an important feature of the invention, a computer logic system (CL) is used to determine the test criteria of the test program. This allows, on the one hand, the consideration of a large number of empirical values ​​regarding possible defects and the probability of their occurrence, and on the other hand, enables the inclusion of interactions between components and defects of different components in the creation of the test program, thus allowing for the creation of better-suited test programs even for more complex products. Consequently, the quality inspection according to the invention allows for an improvement in reliability and / or a reduction in testing effort through better focus and consolidation of the test program.

[0011] According to a preferred embodiment of the invention, event data relating to at least one event that occurred during the processing of the product can be recorded in at least one of the processing steps, and the event data is input into the AI ​​system. Thus, not only statistically possible events, but also events that actually occurred and affected the specific product being tested, can be taken into account by the AI ​​system when creating the test program. Events within the meaning of the embodiment described above refer in particular to series production, in which several identical or similar products are manufactured in larger quantities according to a defined production plan in predefined processing steps.The production plan defines standardized processing steps that are substantially identical (standard) for each manufactured product, and / or standard conditions under which the processing steps are carried out. Events within the meaning of this disclosure are deviations from the production plan, i.e.,

[0012] Deviations from the standard procedure of a processing step, or deviations from the standard conditions during a processing step.

[0013] The event data preferably contains one or more pieces of information selected from:

[0014] - a processing step ID that indicates the processing step in which the event occurs,

[0015] - a product ID that identifies the product being processed in the processing step where the event occurs,

[0016] - a time reference that specifies a point in time or a period of time in which the event occurs,

[0017] - Event flow data that specifies the actual flow of the processing step in which the event occurs or indicates the deviation from the standard flow,

[0018] - Event condition data, which indicates the actual conditions during the processing step or the deviation of the actual conditions from the standard conditions.

[0019] Event data containing this information allows the event to be assigned to the relevant processing step or product, so that the AI ​​system receives information about specific events that occurred in a particular processing step or during the processing of a specific product. In particular, when creating an individual test program for a specific product, the AI ​​can consider event data about an event that occurred on precisely that specific product during a previous processing step. In a further preferred embodiment of the invention, environmental data can be acquired in at least one of the processing steps, indicating the state of the environment of a product processed in that step, and the environmental data is input into the AI ​​system. The environmental data preferably comprises one or more pieces of information selected from:

[0020] - a processing step ID that indicates the processing step in which the environment data is captured,

[0021] - a product ID that identifies the product being processed in the processing step where the environmental data is collected,

[0022] - a time indication that specifies a point in time or a period of time in which the environmental data is collected,

[0023] - Environmental condition data, which specifies the environmental conditions.

[0024] The environmental data, in particular the environmental conditions data, preferably comprise one or more measured values ​​selected from pressure, temperature, and humidity, measured using appropriate sensors, for example, a pressure sensor, a temperature sensor, or a humidity sensor. If the AI ​​system incorporates environmental data into the determination of the test program, especially environmental data that prevailed in the environment during the processing of a specific product, the accuracy of the AI ​​system's prediction of possible errors or tolerance deviations can be improved, thus enabling the creation of a more efficient test program.

[0025] In a further preferred embodiment of the invention, machine data from a machine processing the product to be tested is acquired in at least one of the processing steps, and this machine data is input into the AI ​​system. The use of machine data allows for particular synergies, since such data is already acquired for a large proportion of machines, for example, to control or monitor the machine itself. According to the embodiment of the invention, this data can also be input into the AI ​​system for more efficient creation of the test program. The machine data preferably comprises one or more parameters selected from:

[0026] - a machine ID that identifies the machine or machine section in which the machine data is recorded,

[0027] - a product ID that identifies the product being processed in the machine,

[0028] - a time indication that specifies a point in time or a period of time in which the machine data is recorded,

[0029] - Operating status data, which contains information about the operating status of the machine during the processing of the product.

[0030] The machine data, in particular the operating status data, preferably includes information on the configuration and / or operation of the machine, or includes sensor values ​​that were recorded by sensors of the machine.

[0031] In a further preferred embodiment of the invention, the method comprises at least one intermediate inspection step prior to the product inspection step, in which the product is subjected to a preliminary inspection (intermediate inspection). An intermediate inspection "of the product" also includes an intermediate inspection of a part / component / sub-component of the product or an intermediate inspection of an intermediate product of the product. In this intermediate inspection, an actual value can be determined for at least one product characteristic and compared with a reference value assigned to that product characteristic. The intermediate inspection can result in intermediate inspection data that indicates the result of the aforementioned comparison between the actual value and the reference value and / or the actual value itself.

[0032] The interim inspection data is preferably entered into the AI ​​system to improve the predictive power of the AI ​​system and to create an even more tailored inspection program for the individual product for which the interim inspection data was collected. For this purpose, in addition to the inspection result, the interim inspection data may preferably include one or more of the following information: - an interim inspection ID that identifies the interim inspection performed, and / or a test criterion, and / or a product characteristic tested during the interim inspection.

[0033] - a product ID that identifies the product, or / and part / component / component of the product, or / and intermediate product of the product that was inspected in the intermediate inspection,

[0034] - a time indication specifying a point in time or a period of time during which the intermediate examination was carried out,

[0035] - Environmental condition data, which specifies the environmental conditions at the location of the interim inspection during the interim inspection.

[0036] In a further preferred embodiment of the invention, the specification data of the product to be tested are entered into the quality control system. Additionally or alternatively, order data for the product to be tested can be entered into the quality control system. The order data can, in particular, contain information on a desired configuration or quality of the product. Order data can supplement, modify, or replace the product's specification data. For example, most of the product characteristics can be predefined by the specification data, with some product characteristics being determined by the order data, thus enabling the production of a customized variant of the product. Entering this order data into the quality control system can address specific testing requirements for the product characteristics defined by the order data and thus further improve the efficiency of the quality inspection.

[0037] In a further preferred embodiment of the invention, a test catalog is provided which comprises a set of test criteria for a substantially complete test of the specification data, wherein the test catalog is input to the AI ​​system and wherein the AI ​​system selects the test criteria for the test program from the test criteria of the test catalog. Based on the data input to it, the AI ​​system can then advantageously operate in such a way that, for a specific product or group of products, it determines, for each of the test criteria of the test catalog, a probability that testing the test criterion on the specific product or group of products will lead to a specific result.Based on these probabilities, the method of the exemplary embodiment, in particular the Kl system, can then finally make a selection of the test criteria for the test program, so that, for example, test criteria are selected for the test program for which the probability of a positive test (presence of a defect, a tolerance deviation, etc.) is particularly high.

[0038] In a further preferred embodiment, the test data from completed tests can be fed back to the AI ​​system and re-entered into it as training data. In this way, the AI ​​system receives feedback regarding the actual test result and is further trained by each test of each test criterion (self-learning system). The precision of the test program creation, in particular the precision of predicting the results of tests of specific test criteria, and thus the selection of suitable test criteria for the test program, is thereby improved.

[0039] Embodiments of the invention can achieve a significant increase in the detection rate through the use of causally derived additional tests, which can be context-sensitive and tailored to product characteristics and manufacturing events. At the same time, the testing effort can be reduced through cost-based policy optimization of the test sequences.

[0040] By integrating real-time data from production as well as field feedback, the system in embodiments of the invention can continuously adapt to new error patterns and variants and react flexibly to production deviations.

[0041] In a further preferred embodiment of the invention, at least one test criterion of the test program can contain at least one media file representing a test instruction for testing the product characteristic associated with the test criterion. Such a media file particularly supports manual testing and facilitates the rapid identification of the product characteristic to be tested or the steps to be performed for testing by a test operator, for example, through appropriate audio or video playback. The media file can be an image file or a video file showing a product characteristic of the product to be tested.

[0042] In a further preferred embodiment of the invention, the media file can be created by a generative AI system, wherein the following input is provided to the generative AI system:

[0043] Information on at least one test criterion from the test catalog, and product data containing the specification data and / or at least one product representation file, where the product representation file is an image or video file showing the product or a section of the product. The use of the generative AI system allows for a reduction in data transmission and / or data storage effort, since it is not necessary to provide, store, or transmit separate media data in advance for each possible test instruction or test criterion of the test catalog. Instead, the media file can be generated as needed for a current test instruction or test criterion. Furthermore, the use of a generative AI system allows for a high degree of flexibility in the way the data is displayed, so that the process can be quickly adapted to the user's needs.

[0044] In a further preferred embodiment, the test data from completed tests can be fed back into the generative AI system and re-entered as training data. This training data can be, for example, image or video data created during the testing of product 16a, such as camera recordings of a defective section of product 16a. In this way, the generative AI system is further trained with actual media data of the product (self-learning system). The precision of generating media data for new test criteria and / or new product sections is thus improved.

[0045] In a further preferred embodiment of the invention, the test criteria of the test program can be represented by virtual test cards, which can be displayed to a test operator on a screen of a testing device, wherein exactly one test card is assigned to each test criterion, which shows at least one of the following elements: - the product characteristic assigned to the test criterion,

[0046] - a reference value assigned to the test criterion,

[0047] - at least one media file assigned to the test criterion.

[0048] Virtual test cards of this type allow even complex test programs for manual or semi-automatic testing to be structured, enabling intuitive and reliable execution of the test program by the operator. The testing device can accept input from the operator to display the next test card and / or to record the test data, which supports simple and error-free operation, for example, by advancing through the test cards step by step.

[0049] Alternatively or additionally to manual or semi-automatic testing of test criteria, in a further preferred embodiment of the invention, the testing of the product according to the test criteria can be carried out by an automatic testing device without intervention from a tester. This includes the acquisition and processing of the test data, in particular the return of the test data to the AI ​​system, at least for some of the test criteria. In this way, the speed of the testing process and, if applicable, also the accuracy or reliability of the testing process can be increased.

[0050] According to a second aspect of the invention, the aforementioned problem is solved by a method for training an AI system, wherein the method comprises providing training data, the training data comprising a plurality of training data sets, each training data set comprising a test criterion to which a product feature of a product is assigned, and wherein each training data set further comprises: a reference value and an actual value, both of which are assigned to the product feature, and / or test data indicating the result of a comparison between a reference value and an actual value, both of which are assigned to the product feature; and wherein the training data are input into the AI ​​system. An AI system trained in this way is particularly suitable for use in a method according to the first aspect of the invention in order to achieve the advantages described above in connection with the first aspect of the invention.In a preferred embodiment of the method according to the second aspect of the invention, the training data may further comprise data selected from:.

[0051] - Event data which was recorded in at least one processing step of a plurality of processing steps of a process for manufacturing a product, wherein the event data relates to at least one event that occurred during the processing of the product,

[0052] - Machine data which was recorded in at least one processing step of a plurality of processing steps of a process for manufacturing a product by a machine processing the product,

[0053] - Interim inspection data containing the result of at least one additional intermediate inspection carried out in the manufacturing process prior to the actual inspection in accordance with the inspection criterion,

[0054] - Product specification data, which includes reference values ​​for other product characteristics not assigned to the test criterion,

[0055] - a test catalog that includes a set of test criteria.

[0056] Such enrichment of the training data enables the improvement of the predictive power of the AI ​​system and thus an increase in the efficiency of quality control as well as ultimately an improvement in product quality.

[0057] According to a third aspect of the present invention, the above-mentioned problem is solved by a computer program which, when executed on a computer, is designed to carry out a method according to the first aspect of the invention and / or according to the second aspect of the invention in order to achieve the advantages mentioned above in connection with the first aspect of the invention or the second aspect of the invention.

[0058] Furthermore, according to a fourth aspect, the invention provides a system for testing a product, comprising at least one computer for controlling a method according to the first and / or the second aspect of the invention, in particular a computer on which a computer program according to the third aspect of the invention is installed or installable, wherein the system preferably comprises: a specification data interface for inputting specification data of the product, which contains reference values ​​for product characteristics of the product; a test program module for creating a test program, which comprises a plurality of test criteria, each of which is assigned at least one of the product characteristics; a test module for testing the product according to the test criteria of the test program by testing the product characteristics assigned to the test criteria.wherein reference values ​​assigned to the product characteristics are compared with corresponding actual values ​​of the product characteristics determined during testing, and for recording test data that indicate a result of the test and / or the actual values ​​of the product characteristics, wherein the test program module includes a KL system which is set up to determine the test criteria of the test program.

[0059] The system can be based on a modular software architecture with clearly defined interfaces (APIs) for data exchange between production facilities, AI agents, and testing devices. Data transmission preferably occurs via secure message queues with support for prioritization and persistence. The AI ​​agents preferably run on distributed servers or cloud platforms, with a service-oriented architecture (SOA) ensuring scalability and fault tolerance. Embedded controllers and mobile testing devices communicate with the quality management system via standardized protocols (e.g., OPC UA, MQTT).

[0060] In a method or system according to the invention, an AI system used can be a hierarchically structured, in particular multi-stage, multi-agent system comprising at least one, preferably all of the following specialized agents, which integrate different AI technologies and models:

[0061] Prediction Agent (Level 1): This agent uses an ensemble learning architecture, encompassing deep neural networks (e.g., CNNs, RNNs, LSTMs), and / or gradient boosting machines (e.g., XGBoost, LightGBM), random forests, and / or support vector machines (SVMs). Feature engineering methods and AutoML techniques can also be employed to predict robust failure probabilities for individual product features from heterogeneous data sources (product, process, environmental, and machine data). Time series analysis and / or anomaly detection using variational autoencoders (VAE) and / or isolation forests can complement the prediction to accurately model complex relationships and interactions between components.

[0062] Generation agent (level 2): ​​Building upon the

[0063] This agent generates error probabilities using transformer-based Large Language Models (LLMs, e.g., GPT architectures) and / or Graph Neural Networks (GNNs) representing product structure and 3D CAD data, along with contextualized inspection instructions. Additionally, Natural Language Processing (NLP) techniques such as Named Entity Recognition (NER), Intent Detection, or Text Summarization can be applied to formulate new inspection instructions from unstructured data sources, customer feedback, and process documentation. The output can be in the form of multimodal inspection plans, including visual (images and annotations) and / or textual instructions, which are preferably dynamically adapted to the production environment.

[0064] Optimization Agent (Level 3): This agent implements a combination of deep reinforcement learning (e.g., Proximal Policy Optimization (PPO), Deep Q-Networks (DQN)) and meta-learning approaches (Model-Agnostic Meta-Learning, MAML) to adaptively and resource-efficiently prioritize and optimize inspection sequences. Target parameters can be selected from: minimizing inspection costs, maximizing the defect detection rate, considering safety criticality, and sustainability metrics (e.g., energy consumption, material efficiency). The agent can utilize context-sensitive reward functions and / or continuously learn from real-time production data and / or feedback loops and / or changing quality requirements. Two or more agents can be interconnected via an asynchronous, message-oriented coordination framework, preferably using standardized APIs and message queues (e.g.,MQTT, AMQP) - can capture, prioritize, and integrate intermediate results with low latency in real time. This framework can ensure consistent, coherent decision logic across agents and enable dynamic, adaptive adjustment of testing programs throughout the entire production process.

[0065] In embodiments of the invention, technological advancements can be provided that increase the performance and adaptability of the AI-supported testing program system. These advancements can include, in particular, the use of federated learning in combination with model-agnostic meta-learning (MAML), which allows AI models to be trained jointly and in compliance with data protection regulations across various, geographically distributed production sites. This enables continuous improvement of error prediction without the need to centrally collect sensitive production data.

[0066] In further embodiments of the invention, a feedback agent based on a Large Language Model (LLM) can be used to automatically evaluate and consolidate customer feedback and / or free-text messages from the field. In particular, the feedback / LLM agent can analyze incoming unstructured data using natural language processing (NLP) and extract relevant topics, fault descriptions, and / or suggestions for improvement. Techniques such as intent recognition, named entity recognition, and / or context-based text classification can be employed. The insights gained can be automatically translated into new or modified test criteria, enabling the system to continuously learn and allowing test programs to be improved based on real customer feedback.The feedback agent can, in particular, independently generate new testing knowledge and thus adaptively and contextually expand the testing program library. Natural language processing (NLP) can enable the use of unstructured data sources that have previously received little attention in quality assurance systems. In a further embodiment of the invention, a completeness check of the test catalog can be implemented using a CAD-supported mapping method. Here, preferably 2D images of the product are compared with 3D CAD models to ensure that all relevant product areas are covered by testing criteria and, in particular, that no testing gaps arise.

[0067] In a further embodiment of the invention, the system can include or use Causal AI and / or the method can use Causal AI, thus enabling a causal-based analysis. Here, text modules and process data are preferably evaluated to automatically generate test content that appears meaningful due to causal relationships, e.g., in the case of a rear seat defect, the inspection of surrounding components or the interior environment.

[0068] A further feature of embodiments of the invention can be the consideration of sustainability goals in the selection and weighting of test criteria. Sustainability assessment can be achieved, in particular, by integrating quantifiable environmental metrics, such as material efficiency, energy consumption, and CO2 footprint, into the optimization process. The AI ​​system can evaluate the impact of individual test criteria on these metrics and prioritize tests that promote a reduction in resource consumption. In this way, test programs can be created that pursue ecological sustainability goals in addition to defect detection, without compromising product quality or safety.

[0069] The system can integrate both structured data, such as sensor readings, machine data, and event data, as well as unstructured data, e.g., from NLP-analyzed error reports. This enables more comprehensive and precise error detection.

[0070] Finally, embodiments of the invention can provide for real-time adaptation of the test programs, enabling an immediate response to new events and deviations during the ongoing production process. This real-time adaptation can be achieved, in particular, through continuous monitoring of relevant production data, which can automatically initiate an update of the test program when defined thresholds are exceeded. The AI ​​agents can evaluate this trigger data, check existing test plans for accuracy, and, if necessary, generate modified test criteria, which are preferably fed directly into the test process. To avoid inconsistencies, a synchronization mechanism can also be used to coordinate parallel test sequences and detect potential conflicts.Additionally, supplier data can be included in the analysis to identify and address systematic sources of error at an early stage.

[0071] In contrast to the prior art, embodiments of the invention can generate individually contextualized test content for each product, rather than simply selecting from predefined test categories. While previously only a selection from test categories based on binary decisions (test or don't test) or a limited number of test classes (functional, noise, or visual inspection) was possible, embodiments of the present invention can generate granular, product-specific test plans with detailed test content. Such test content goes far beyond the selection of predefined test categories. This allows for significantly more precise and resource-efficient quality control, tailored to the actual error probabilities arising from product configuration and process history.

[0072] In particular, the test program created according to the invention preferably comprises a new combination of test criteria, which as such a combination was not stored in an existing database of the system or in any other existing data set to which the method or the system of the invention has access. In particular, when creating the test program, the Kl system preferably makes a genuine selection of k test criteria from a set of n test criteria, where k is less than n, and where information about which test criteria were selected, i.e., the specific test program created, was not present in any data set of the system prior to the step of creating the test program. The invention is explained in more detail below with reference to an exemplary embodiment and the accompanying drawing.

[0073] Figure 1 shows a schematic representation of a production plant for manufacturing a product, in which a process according to the exemplary embodiment takes place.

[0074] A production plant 10 shown in Figure 1 can, for example, comprise a production line 12 along which a plurality of processing stations 14a, 14b, ... are provided, in which a product 16 to be manufactured is successively subjected to various processing steps. The processing steps can each involve the machining of an intermediate product, the assembly of components of the product, a transport or transfer operation, or other handling or processing of the product or its intermediate product. For example, the production line 12 can be a vehicle production line for the manufacture of vehicles, comprising a first processing station 14a for painting a body (intermediate product) of the vehicle and a second processing station 14b for assembling a vehicle seat.

[0075] Naturally, additional processing stations will usually be provided, which are not shown in the example drawing.

[0076] After processing, and in particular after leaving the last processing station 14b of the production line 12, product 16 passes through an inspection station 18 where a quality inspection takes place. During this inspection, the product characteristics of product 16 are determined and compared with corresponding reference values ​​to detect any defects in the manufactured product and / or deviations from the product's specifications or order data. The inspection can be carried out manually by an inspector 20, automatically by an electronic inspection device 22, or semi-automatically through cooperation between the inspector 20 and the inspection device 22.Production facility 10 also includes a QM system 24 (quality management system), which comprises one or more interconnected computers and connectivity means for input and output of data and / or for the exchange of data with production line 12, with other local or remote computers or with a network, in particular the Internet.

[0077] The QM system 24 can have a product recording device 26 which is set up to uniquely record a product 16 currently being processed in a specific processing station 14a, 14b and, for example, to read in a unique product ID of the product 16.

[0078] Furthermore, the QM system 24 can include an event recording device 28, which is configured to record the occurrence of an event during the processing of a specific product 16 at a specific processing station 14a, 14b in the form of event data, i.e., deviations from a production plan or deviations from standard conditions specified for the planned processing of product 14. If, as in the example mentioned above, the second processing station 14b is configured for assembling a seat, the event recording device 28 can be configured to receive data from a process control system of the production plant 10 indicating that a first seat was installed and, after a defect was detected, removed and replaced with a second seat, thus resulting in an additional seat removal and assembly compared to the production plan.The additional seat disassembly and seat assembly then constitute an event within the meaning of the present invention.

[0079] The quality management system may also include an environmental monitoring device 30, which is configured to indicate the condition of an environment in or around the processing station, i.e., the condition of the environment of the product 16 being processed in the processing station, in the form of environmental data. The environmental monitoring device 30 may include at least one environmental sensor. If, as in the example mentioned above, the first processing station 14a is configured for painting a car body, the environmental monitoring device 30 may include a temperature sensor and / or a pressure sensor to monitor compliance with standard temperature and / or pressure conditions specified for painting.

[0080] The data acquisition devices 26, 28, 30 are shown in Fig. 1 near the processing stations 14a, 14b for illustrative purposes only, to symbolize a logical assignment. They can be physically or functionally part of the processing stations 14a, 14b, but can also be located away from the processing stations, for example in other parts of the QM system 24, and in particular be functionally implemented as a computer program on a computer of the QM system 24.

[0081] The QM system 24 preferably further comprises a machine data interface for inputting machine data from the processing stations 14a and 14b, which contains information on the current operating states and operating conditions of the processing stations 14a and 14b. In addition, the QM system 24 preferably accesses specification data containing reference values ​​for product characteristics of the series production of product 16, and / or order data specifying a desired configuration variant or special characteristic of product 16, i.e., a deviation from or supplement to the specification data desired individually for a specific product 16.

[0082] The QM system 24 comprises a first AI system 40, for example with a neural network, which is configured and specifically trained to support an increase in the efficiency of quality control, and for this purpose, for example, to calculate a prediction for the occurrence of a defect in the manufactured product 16 and / or a deviation of the manufactured product 16 from specification data or order data of the product 16. The first AI system 40 can receive the following as input data:

[0083] - a product ID provided by the product acquisition device 26 of a product 16 currently being processed in a specific processing station 14a, 14b, and / or - data from the event acquisition device 28 concerning the occurrence of an event during the processing of the product 16 identified by the product ID, and / or

[0084] - Environmental data provided by the environmental sensing device 30 for the product identified by the product ID 16 and / or

[0085] - Machine data entered via machine data interface 32 for processing station 14a or 14b, in which product 16 identified by the product ID is or was processed, and / or

[0086] - Interim audit data, in particular interim audit results, of upstream interim audits (not shown in the figure)

[0087] - the specification data and / or

[0088] - the order data.

[0089] Based on an output from the first quality control system 40, the quality management system 24 can then determine a preselection of one or more test criteria 42, which are selected from a test catalog 44 entered into the quality management system 24. This catalog contains all quality control measures that can be meaningfully performed on the product 16 for a complete inspection of the product characteristics of the product 16. For this preselection, those test criteria 42 are preferentially chosen for which the probability of a positive defect / deviation test (defect or deviation present) is particularly high. Test criteria that have special priority for compliance reasons, due to special safety standards, or due to a customer request are also given preferential or even mandatory consideration in the preselection.

[0090] The test catalog 44 can be configured such that the test criteria 42 contained therein, which are intended for manual or semi-automatic testing, each include test instructions in text form for later transmission to the tester 20. The test instructions can then be displayed on a mobile testing device 45. In the preferred embodiment shown in Fig. 1, however, the test criteria 42 are input into a generative AI system 46, which also receives as further input data the specification data and / or the order data and / or media data (in particular image data) of the product 16 or of sections of the product 16.The generative AI system 46 can contain a neural network trained on historical training data, which is configured or trained to generate a media file 50 for each entered test criterion based on the input data. This media file represents a visual inspection instruction for testing the product characteristic associated with the test criterion. The media file generated for a test criterion can preferably be in the form of a test card 50a, 50b, ... so that test criteria can be visualized and handled electronically at a later time as a stack of test cards 50a, 50b, ...

[0091] The QM system 24 preferably comprises a second KL system 52, which is configured to select, from the input test criteria 42, in particular the pre-selection test criteria 42, precisely those test criteria actually to be tested for a specific product 16 in the test station 18 and to combine them into a test program 54. Safety data 58, containing information about the influence of a product characteristic assigned to a predetermined test criterion on the safety of the product 16, can be input into the second KL system 52 as further input data. The second KL system 52 can then be trained such that, when selecting the test criteria for the test program 54, it gives greater weight to test criteria relating to safety-critical product characteristics and tends to include them in the selection even if the probability of a positive test result is lower.

[0092] Alternatively or additionally, cost data can be entered into the second AI system 52, indicating the effort required to test a predetermined test criterion. The second AI system 52 can be trained such that, when selecting test criteria for the test program 54, it gives lower weight to test criteria that involve costly tests but do not concern safety-critical product characteristics, and even if a positive test result is more likely, it tends not to include them in the selection. Further data that can be entered into the AI ​​system in a method according to the invention, for example, the second AI system 52, are:

[0093] - Intermediate inspection data, which contains the result of an intermediate inspection of product 16a, which was carried out in one of the intermediate inspection stations upstream of the inspection station 18 along the production line 12 and which is not shown in the figure, or complaint effort data, which represents an effort associated with the processing of a complaint that is triggered by a product characteristic assigned to a predetermined inspection criterion showing a defect or a deviation from the specification and / or order data of the product.

[0094] The test program 54 generated by the second AI system 52 can then be transmitted to the test station 18 and, in particular, displayed on a screen of a test device 45. The test device 45 is preferably a mobile device, for example, a smartphone or a tablet computer, which can be operated by the tester to carry out the tests. Preferably, the test criteria of the test program 54 are displayed on the test device 45 in the form of the test cards 50a, 50b, ... described above. The tester 20 can then process the test criteria by advancing the test cards 50a, 50b individually and receiving the respective test instructions based on the displayed media files. Alternatively, the test criteria of the test program 54 can be transmitted to the test equipment 22 of the test station for automatic or semi-automatic testing.

[0095] In test station 18, the tests are carried out according to test program 54, and test data is recorded. This data specifies a test result and / or an actual value of the product characteristic assigned to each test criterion for each test criterion. In a simple version, the test data can contain binary information (test flag) (error yes / no). Preferably, however, the test data contains further information selected from:

[0096] - a product ID that identifies the tested product,

[0097] - a time indication specifying the time or period of testing, an actual value of the product characteristic assigned to the test criterion.

[0098] The production plant 10 uses the recorded test data in a manner known per se for further control of the production process, in particular for sorting out defective products, for rectifying errors, for readjusting processing stations 14a, 14b, ... , towards reducing deviations between actual values ​​of product characteristics and their target values ​​according to the specification and / or order data, or for reporting to a QM management system .

[0099] In the embodiment of the invention, the QM system 24 can further utilize the test data or data corresponding to the test data as initial training data 56 for the first AI system 40 and / or as initial training data 58 for the second AI system 52, so that the AI ​​systems can be (further) trained during the use of the QM system 24 in ongoing production operations. For example, by comparing the test data with the preselection of test criteria 42 made by the first AI system 40, a reduction in the predictive power of the first AI system 40 can be detected, which can be used as a trigger for a training measure using the training data 56 (model decay-based retraining). Similarly, the training data 58 can be used by an evaluation module 60 to evaluate the predictive power of the second AI system 52, and the result of the evaluation can be input to the second AI system 52.The second Kl-System 52 can use the result to update its parameters or boundary conditions, or directly feed it as a further input value to a neural network of the second Kl-System 52.

[0100] Test data or other data recorded during an inspection at test station 18 can also be entered into the generative AI system 46 as third-party training data 62, in particular to improve the representation of the test criteria in the media file and to ensure fast, intuitive, and / or reliable recognition of the test criteria or the product characteristics to be inspected by the inspector 20. For example, image or video data recorded by a camera during the inspection can be supplied to the generative AI system 46 as additional training data 62 (live training data). Test data from test station 18 can also be used by the QM system 24 to adapt the test catalog 44, in particular to modify test criteria and / or to include new test criteria in the test catalog 44 and / or to remove existing test criteria from the test catalog 44.Furthermore, it is conceivable that the test data could be used to capture live trends and correlations, which could then be used to improve the prediction of defects or deviations, enabling reactive improvements in the manufacturing process. In another scenario, such live trends and correlations could be used to support the testing process itself at test station 18, for example, by controlling the computer vision systems at test station 18. Additionally, the captured live trends and correlations could be used for supplier analysis, for example, to establish relationships between specific suppliers and specific systematic defects or deviations, and to take these relationships into account when creating test programs.Test data or live trends and correlations derived from it can also be used for faster field analysis, for example to analyze which process steps in the field lead to systematic problems.

[0101] In a production plant 10 of the exemplary embodiment, a method for manufacturing a product 16 according to an exemplary embodiment of the present invention can proceed as follows.

[0102] In series production, several products 16 are manufactured according to the same specification data and / or order data. The production and testing of an individual product 16a from the series is now considered.

[0103] Product 16a passes through various processing stations 14a, 14b, ... along production line 12, where event data and / or environmental data can be determined at each processing station. These data relate to the processing of the individual product 16a at the respective processing station 14a, 14b, ... (i.e., contain or allow a direct or indirect link to the individual product 16a and / or the respective processing station 14a, 14b, ...). The data is preferably input to the first AI system 40, so that after all processing stations 14a, 14b, ... have passed through the first AI system 40, a data record containing event data and / or environmental data for each processing station 14a, 14b for the individual product 16a.From this data, the QM system 24, using the first KL system 40 and / or the second KL system 52, can generate the individual test program 54 for the individual product 16a in the manner described above, i.e., a selection of test criteria that can be displayed and managed in the form of test cards on a screen of a test device 45.

[0104] In test station 18 (or in connection with test station 18), the tests are carried out manually, semi-automatically, or fully automatically according to the test criteria of test program 54, and the test results, in particular the occurrence of defects or deviations of product characteristics from the specification and / or order data, are recorded in the form of test data. The test data then refers to the specific product 16a and takes into account the individual history of product 16a during its manufacture in processing stations 14a, 14b, ... of production line 12, i.e., the respective events or environmental conditions that were recorded for precisely this specific product 16a.

[0105] The test data or data derived from it are ultimately fed back to the first and / or second AI system as training data in order to implement continuous training of the first and / or second AI system and to improve the predictive power or precision of the first and / or second AI system.

Claims

Claims 1. A method for manufacturing a product, comprising: a. providing specification data of the product which includes reference values ​​for product characteristics of the product; b. processing the product in a plurality of processing steps according to the specification data; c. creating a test program which includes a plurality of test criteria, each of which is assigned at least one of the product characteristics; d. testing the product according to the test criteria of the test program, by testing the product characteristics assigned to the test criteria, comparing the reference values ​​assigned to the product characteristics with the corresponding actual values ​​of the product characteristics determined during the test; and e.Recording test data that indicates a result of the test and / or the actual values ​​of the product characteristics, characterized in that in the step of creating the test program the test criteria of the test program are determined using a KL system.

2. Method according to claim 1, wherein in at least one of the processing steps event data is recorded which relates to at least one event that occurred during the processing of the product, and that the event data is entered into the AI ​​system.

3. Method according to claim 1 or claim 2, wherein in at least one of the processing steps machine data of a machine processing the product to be tested are recorded, and the machine data are entered into the AI ​​system.

4. A method according to at least one of the preceding claims, wherein the method comprises at least one intermediate testing step prior to the step of testing the product, in which the product is subjected to an intermediate test, - wherein in the interim inspection an actual value is determined for at least one product characteristic of the product and compared with a reference value assigned to the product characteristic, - wherein, as a result of the interim audit, interim audit data are recorded, indicating the result of the comparison between the actual value and the reference value and / or the actual value itself, and - whereby the intermediate examination data is entered into the KL system.

5. Method according to at least one of the preceding claims, wherein the specification data is input into the AI ​​system.

6. Method according to at least one of the preceding claims, wherein a test catalog is provided which includes a set of test criteria for a substantially complete test of the specification data, wherein the test catalog is input to the AI ​​system and wherein the AI ​​system selects the test criteria for the test program from the test criteria of the test catalog.

7. Method according to at least one of the preceding claims, wherein the test data of completed tests are fed back to the AI ​​system and re-entered into the AI ​​system as learning data.

8. Method according to at least one of the preceding claims, wherein at least one test criterion of the test program contains at least one media file which represents a test instruction for testing the product feature assigned to the test criterion.

9. Method according to claim 8, wherein the media file is an image file or a video file which shows a product feature of the product to be tested.

10. Method according to claim 8 or claim 9, wherein the media file is created by a generative AI system, wherein the following is input to the generative AI system: - at least one test criterion from the test catalog, and - Product data, which includes the specification data and / or at least one product representation file, where the product representation file is an image or video file showing the product or a portion of the product.

11. Method according to at least one of the preceding claims, wherein the test criteria of the test program are represented by virtual test cards which can be displayed to a test person on a screen of a test device, wherein each test criterion is assigned exactly one test card which shows at least one of the following elements: - the product characteristic assigned to the test criterion, - a reference value assigned to the test criterion, - at least one media file assigned to the test criterion.

12. Method according to claim 11, wherein the testing device receives input from the tester to display the next test card and / or to record the test data.

13. Method according to at least one of the preceding claims, wherein the testing of the product according to the test criteria, the recording of the test data and the processing of the test data, in particular the return of the test data to the AI ​​system, is carried out by an automatic testing device without intervention by a test person, at least for some of the test criteria.

14. Procedure for training an AI system, comprehensive: - Providing training data, comprising a plurality of training datasets, wherein each training dataset includes a test criterion to which a product feature of a product is assigned, and wherein each training dataset further includes: o a reference value and an actual value, both of which are assigned to the product characteristic, and / or o test data, which indicate the result of a comparison between a reference value and an actual value, both of which are assigned to the product characteristic, - Entering the training data into the AI ​​system.

15. The method of claim 14, wherein the training data further comprises: - Event data which was recorded in at least one processing step of a plurality of processing steps of a process for manufacturing a product, wherein the event data relates to at least one event that occurred during the processing of the product, - Machine data which was recorded in at least one processing step of a plurality of processing steps of a process for manufacturing a product by a machine processing the product, - Interim audit data, in particular interim audit results, of upstream interim audits - Product specification data, which includes reference values ​​for other product characteristics not assigned to the test criterion, - a test catalog that includes a set of test criteria.

16. Computer program which, when executed on a computer, is designed to perform a method according to one of the preceding claims.

Images