DE102021207993A1 - Method for training a machine learning algorithm for determining an actual state of at least one component property of a heat-treated metallic component - Google Patents

Abstract

The invention relates to a method for training a machine learning algorithm for determining an actual state of at least one component property of a heat-treated metallic component, the method (1) having the following steps: During the heat treatment of a metallic component, recording data characterizing the heat treatment (2), determining the actual state of the at least one component property from the recorded data (3), and training the machine learning algorithm based on the recorded data and the actual state of the at least one component property features ( 4).

Description

The invention relates to a method for training a machine learning algorithm for determining an actual state of at least one component property of a heat-treated metallic component and a method for quality assurance of heat-treated metallic components based on the machine learning algorithm for determining an actual state of at least a component property of a heat-treated metallic component with which a quick, effective and complete inspection of the quality of heat-treated metallic components can be achieved in a simple manner.

Heat treatment is generally understood to mean processes for the treatment of workpieces or metallic components, in which heating and cooling are carried out in a controlled manner in order to change the material properties. Examples of such heat treatment processes are annealing treatments, tempering treatments or hardening processes.

Heat treatment methods of this type also influence, among other things, an edge zone of the metallic component, the formation of the edge zone and its properties having a decisive importance for the usage behavior and useful life of the metallic component. Since conventional heat treatment processes usually do not allow any statements to be made about the final state of the surface zone of the heat-treated component, a final, precise examination of the heat-treated surface zone must usually be carried out after the heat treatment has taken place in order to make statements about the component properties such as hardening depth, hardness, microstructure, etc. to be able to meet, and thus about the usage behavior and service life of the corresponding heat-treated component.

It is known, for the final assessment of the heat-treated component and in particular the edge zone of the heat-treated component, to subject this to a visual check, which, however, has only a limited, subjective significance and is restricted to surface areas. For a further qualitative assessment of the heat-treated component, metallographic observations of the microstructure and hardness measurements in the section as well as residual stress and residual austenite measurements can be carried out using an X-ray diffractometer, although this sometimes requires the component to be destroyed. Test methods based on micromagnetic measuring methods also allow comprehensive and non-destructive testing of the surface zone properties of heat-treated components and the quantitative determination of various material parameters, but only provide locally valid information and are limited to magnetizable materials.

From the pamphlet WO 2015/169309 A1 a method for quality assurance of at least one component during its production is known, whereby the production takes place by means of an additive manufacturing method with at least one processing laser, whereby the component is built up in layers, thermographically a large number of images, in a defined period of time, of at least one component area on the laser beam are recorded by means of at least one recording sensor and changes over time in the heat distribution in a component area free of a molten pool are recorded, with the occurrence of a fault in the top component layer or below being detected using a characteristic change over time in the heat distribution at the fault.

The invention is therefore based on the object of enabling improved testing of the quality of a heat-treated component.

This object is achieved with a method for training a machine learning algorithm for determining an actual state of at least one component property of a heat-treated metallic component according to the features of patent claim 1 .

This object is also achieved by a method for quality assurance of heat-treated metallic components according to the features of patent claim 5.

This object is also achieved by a control unit for training a machine learning algorithm for determining an actual state of at least one component property of a heat-treated metal component with the features of patent claim 7 and a control unit for quality assurance of heat-treated metal components with the features of patent claim 12 .

Advantageous embodiments and developments result from the dependent claims and from the description with reference to the figures.

Disclosure of Invention

According to one embodiment of the invention, this object is achieved by a method for training a machine learning algorithm for determining an actual state of at least one component property of a heat treatment unit th metallic component, wherein, during the heat treatment of the metallic component, the heat treatment characterizing data are recorded, the actual state of the at least one component property characterizing features are determined from the recorded data, and a machine learning algorithm based on the recorded data and the , the actual state of the data characterizing at least one component property is trained.

The data characterizing the heat treatment is understood to mean the data characterizing the metal component heat-treated by a heat treatment device or data characterizing the operation of the heat treatment device during the heat treatment of the metal component. The data characterizing the heat treatment can be obtained by monitoring the metallic component and/or the heat treatment device during the heat treatment, for example by means of sensors.

The actual state of the data characterizing at least one component property is further understood to mean features or statistical models or characteristic values in the time and/or frequency domain generated from the data characterizing the heat treatment by data processing.

Algorithms for machine learning are also based on the fact that statistical methods are used to train a data processing system in such a way that it can perform a specific task without originally being programmed explicitly for this purpose. The goal of machine learning is to construct algorithms that can learn from data and make predictions. These algorithms create mathematical models that can be used, for example, to classify data.

The machine learning algorithm for determining an actual state of at least one component property of a heat-treated metallic component can thus be used to achieve a quick, effective and complete inspection of the quality of heat-treated metallic components in a simple manner. The machine learning algorithm can be used to check the quality of heat-treated components in almost real time and especially immediately after the heat treatment has been carried out, without the need for complex testing methods following the heat treatment, such as a visual check, metallographic microstructure observations and hardness measurements in the section as well as Residual stress and residual austenite measurements using an X-ray diffractometer, or test methods based on micromagnetic measuring methods would be required. During quality inspection, poor-quality heat-treated components that may pose a safety risk can then be sorted out without having to exclude the entire batch of heat-treated components at the same time, since each component can be individually inspected based on the machine learning algorithm. Overall, based on the machine learning algorithm for determining an actual state of at least one component property of a heat-treated metallic component, improved testing of the quality of a heat-treated component is made possible.

The recorded data characterizing the heat treatment can be sensor data or data recorded by a sensor.

A sensor, which is also referred to as a detector, (measuring variable or measuring) recorder or (measuring) sensor, is a technical component that qualitatively or quantitatively records certain physical or chemical properties and/or the material composition of its environment can.

In this way, the heat treatment-characterizing conditions outside the actual data processing system, on which the machine learning algorithm for determining an actual state of at least one component property of a heat-treated metallic component is trained, can be recorded and during the training of the machine learning algorithm are taken into account. Furthermore, data characterizing the heat treatment obtained in a different way can also be recorded and taken into account when training the machine learning algorithm.

The sensor data can include structure-borne noise signals, vibration signals and data about a current electrical output, a current current intensity and/or a current applied voltage of a corresponding heat treatment device. Other signals can be, for example, temperature, pressure, forces, strain, torque, etc.

Structure-borne noise is the sound that propagates in a solid body, with the recording and analysis of structure-borne noise, for example, to detect (stress) cracks in the heat-treated metallic component or for detecting a hardening depth.

Based on the vibration signals, which can be detected by a vibration sensor, vibrations in the heat-treated component can be detected, as a result of which, for example, conclusions can be drawn about the microstructure of the heat-treated metallic component.

Based on the current electrical power, the current current intensity and/or the current voltage applied to the corresponding heat treatment device, conclusions can also be drawn about the operation of the heat treatment device, for example positions of a tool spindle. The corresponding heat treatment device is understood here to mean the device which was used for the heat treatment of the metallic component.

However, the fact that the sensor signals include structure-borne noise signals, vibration signals and data about a current electrical output, a current current intensity and/or a current voltage of a corresponding heat treatment device is only a preferred embodiment. Furthermore, the sensor data can also act on other signals detected by sensors, from which conclusions can be drawn about properties of the heat-treated component or faults in the heat-treated component.

Overall, a machine learning algorithm can be trained based on data characteristic of the heat treatment recorded during a heat treatment of a first part of a batch of metallic components, based on which the remaining part of the metallic components of the batch can then be subjected to a quality check. Furthermore, if multiple metals are to be heat treated, a separate machine learning algorithm can also be trained for each metal.

In one embodiment, the method further includes a step of extracting non-significant features that characterize the actual state of the at least one component property before training the machine learning algorithm. Due to the fact that non-significant features or statistical characteristics are extracted or cut out before the machine learning algorithm is trained, the machine learning algorithm training can be simplified and accelerated while the information content remains the same.

Features or static characteristics that are not meaningful are understood to mean features or static characteristics that are not meaningful for the actual state of the at least one component property or have little or no influence on the training of the machine learning algorithm.

This can also make it possible for the method to be executed on control devices that have comparatively low storage or computing capacities, for example a control device that is integrated into a heat treatment device itself.

A further embodiment of the invention also specifies a method for quality assurance of heat-treated metallic components, in which a metallic component is heat-treated by a heat treatment device in order to achieve a desired target state of at least one component property, with an actual state of the at least one component property a machine learning algorithm trained by an above-described method for training a machine learning algorithm for determining an actual state of at least one component property of a heat-treated metallic component for determining an actual state of at least one component property of a heat-treated metallic component is determined, wherein the Actual state of the at least one component property is compared with the desired target state, and the component is discarded if the actual state of the at least one component property deviates from the target state by more than a permissible threshold value.

In this case, the target state is understood to mean the state of the at least one component property of the heat-treated body, which is to be achieved or set by heat treatment.

The fact that the component is discarded also means that the component is sorted out and usually disposed of.

A method is thus specified with which an improved inspection of the quality of a heat-treated component is made possible. In particular, the machine learning algorithm for determining an actual state of at least one component property of a heat-treated metallic component can be used for this purpose be used to easily achieve fast, effective and complete quality testing of heat treated metallic components. The algorithm of machine learning can be applied to do this in near real time. In particular, to check the quality of heat-treated components immediately after the heat treatment has been carried out, without the need for complex testing methods following the heat treatment, such as a visual inspection, metallographic microstructure observations and hardness measurements in the section, as well as residual stress and retained austenite measurements using an X-ray diffractometer, or testing methods based on micromagnetic measuring methods would be. During quality inspection, poor-quality heat-treated components that may pose a safety risk can then be sorted out without having to exclude the entire batch of heat-treated components at the same time, since each component can be individually inspected based on the machine learning algorithm.

The method can further include a step of adapting operating parameters of the heat treatment device based on a deviation of the actual state of the at least one component property from the target state in such a way that the actual state of the at least one component property during subsequent heat treatments of metallic components by the heat treatment device should match the target state again if the actual state of the at least one component property deviates from the target state by more than a permissible threshold value. This intervention in the heat treatment process or this adjustment of operating parameters of the heat treatment device can significantly increase the effectiveness and productivity of the heat treatment, in particular if a complete batch of metallic components is to be heat treated.

A further embodiment of the invention also specifies a control unit for training a machine learning algorithm for determining an actual state of at least one component property of a heat-treated metal component, the control unit having a receiving unit which is designed during the heat treatment of a metal component to capture data characterizing the heat treatment, a determination unit, which is designed to determine the actual state of the at least one component property from the captured data, and a training unit, which is designed to use the machine learning algorithm based on the captured data and to train the actual state of the at least one component property characterizing features.

A control unit is thus specified with which a machine learning algorithm can be trained, with which improved testing of the quality of a heat-treated component is made possible. In particular, the machine learning algorithm for determining an actual state of at least one component property of a heat-treated metallic component can be used to achieve a quick, effective and complete inspection of the quality of heat-treated metallic components in a simple manner. The machine learning algorithm can be used to check the quality of heat-treated components in near real-time, especially immediately after the heat treatment, without the need for complex testing methods that follow the heat treatment, such as a visual check, metallographic microstructure observations and hardness measurements in the section, as well as residual stress - and retained austenite measurements using an X-ray diffractometer, or test methods based on micromagnetic measuring methods would be required. During quality inspection, poor-quality heat-treated components that may pose a safety risk can then be sorted out without having to exclude the entire batch of heat-treated components at the same time, since each component can be individually inspected based on the machine learning algorithm.

The recorded data characterizing the heat treatment can in turn be sensor data. Thus, in a simple way, the heat treatment-characterizing conditions outside of the actual data processing system, on which the machine learning algorithm for determining an actual state of at least one component property of a heat-treated metallic component is trained, is detected and during training of the algorithm of the machine learning are taken into account. Furthermore, however, data obtained in a different way and characterizing the heat treatment can also be taken into account when training the algorithm of the machine learning.

The sensor data can in turn structure-borne noise signals, vibration signals and data on a current electrical power, a current current and / or a current voltage corresponding heat treatment device include.

However, the fact that the sensor signals include structure-borne noise signals, vibration signals and data about a current electrical output, a current current intensity and/or a current voltage of a corresponding heat treatment device is only a preferred embodiment. Furthermore, the sensor data can also act on other signals detected by sensors, from which conclusions can be drawn about properties of the heat-treated component or faults in the heat-treated component.

In one specific embodiment, the control unit also has an extraction unit, which is designed to extract non-descriptive features that characterize the actual state of the at least one component property before the machine learning algorithm is trained. Due to the fact that non-significant features or statistical characteristics are extracted or cut out before the machine learning algorithm is trained, the machine learning algorithm training can be simplified and accelerated while the information content remains the same. This can also make it possible for the machine learning algorithm to be trained on control devices that have comparatively low storage or computing capacities, for example a control device that is integrated into a heat treatment device itself.

A further embodiment of the invention also specifies a system for training a machine learning algorithm for determining an actual state of at least one component property of a heat-treated metallic component, the system having at least one sensor for detecting data characterizing a heat treatment of the metallic component during the heat treatment of the metallic component and a control device described above for training a machine learning algorithm for determining an actual state of at least one component property of the heat-treated metallic component.

A system is thus specified with which a machine learning algorithm can be trained, with which improved testing of the quality of a heat-treated component is made possible. In particular, the machine learning algorithm for determining an actual state of at least one component property of a heat-treated metallic component can be used to achieve a quick, effective and complete inspection of the quality of heat-treated metallic components in a simple manner. The machine learning algorithm can be used to check the quality of heat-treated components in near real-time, especially immediately after the heat treatment, without the need for complex testing methods that follow the heat treatment, such as a visual check, metallographic microstructure observations and hardness measurements in the section, as well as residual stress - and retained austenite measurements using an X-ray diffractometer, or test methods based on micromagnetic measuring methods would be required. During quality inspection, poor-quality heat-treated components that may pose a safety risk can then be sorted out without having to exclude the entire batch of heat-treated components at the same time, since each component can be individually inspected based on the machine learning algorithm.

A further embodiment of the invention also specifies a control unit for quality assurance of heat-treated metallic components, the control unit having a receiving unit which is designed to have a control unit described above for training a machine learning algorithm for determining an actual state of at least a machine learning algorithm trained on a component property of a heat-treated metallic component to determine an actual state of at least one component property of a heat-treated metallic component, a determination unit which is designed to receive an actual state of the at least one component property of a metallic component heat-treated by a heat treatment device based on the machine learning algorithm for determining an actual state of at least one component property of a heat-treated metal To determine the component, a comparator, which is designed to compare the determined actual state of the at least one component property with a desired target state, and an output unit, which is designed to output that the metallic component should be discarded if the actual - the state of the at least one component property deviates from the target state by more than a permissible threshold value.

A control device is thus specified, with which improved testing of the quality of a heat-treated component is made possible. In particular, the machine learning algorithm for determining an actual state of at least one component property of a heat-treated metallic component can be used to achieve a quick, effective and complete inspection of the quality of heat-treated metallic components in a simple manner. The machine learning algorithm can be used to check the quality of heat-treated components in almost real time, especially immediately after the heat treatment has been carried out, without the need for complex testing methods following the heat treatment, such as a visual check, metallographic microstructure observations and hardness measurements in the section as well as Residual stress and residual austenite measurements using an X-ray diffractometer, or test methods based on micromagnetic measuring methods would be required. During quality inspection, poor-quality heat-treated components that may pose a safety risk can then be sorted out without having to exclude the entire batch of heat-treated components at the same time, since each component can be individually inspected based on the machine learning algorithm.

A further embodiment of the invention also specifies a system for quality assurance of heat-treated metallic components, the system having a heat treatment device for heat-treating metallic components and a control device described above for quality assurance of heat-treated metallic components.

A system is thus specified with which improved testing of the quality of a heat-treated component is made possible. In particular, the machine learning algorithm for determining an actual state of at least one component property of a heat-treated metallic component can be used to achieve a quick, effective and complete inspection of the quality of heat-treated metallic components in a simple manner. The machine learning algorithm can be used to check the quality of heat-treated components in almost real time, especially immediately after the heat treatment has been carried out, without the need for complex testing methods following the heat treatment, such as a visual check, metallographic microstructure observations and hardness measurements in the section as well as Residual stress and residual austenite measurements using an X-ray diffractometer, or test methods based on micromagnetic measuring methods would be required. During quality inspection, poor-quality heat-treated components that may pose a safety risk can then be sorted out without having to exclude the entire batch of heat-treated components at the same time, since each component can be individually inspected based on the machine learning algorithm.

The control unit for quality assurance of heat-treated metallic components can also initiate an adjustment unit, which is designed to initiate operating parameters of the heat treatment device based on a Deviation of the actual state of the at least one component property can be adjusted from the target state in such a way that the actual state of the at least one component property should match the target state again during subsequent heat treatments of metallic components by the heat treatment device. This intervention in the heat treatment process or this adjustment of operating parameters of the heat treatment device can significantly increase the effectiveness and productivity of the heat treatment, in particular if a complete batch of metallic components is to be heat treated.

In summary, it can be stated that the present invention enables improved testing of the quality of a heat-treated component.

The configurations and developments described can be combined with one another as desired.

Further possible configurations, developments and implementations of the invention also include combinations of features of the invention described above or below with regard to the exemplary embodiments that are not explicitly mentioned.

character list

The accompanying drawings are provided to provide a further understanding of embodiments of the invention. They illustrate embodiments and, together with the description, serve to explain principles and concepts of the invention.

Other embodiments and many of the foregoing advantages will become apparent by reference to the drawings. The illustrated elements of the drawings are not necessarily shown to scale with respect to one another.

• 1 ein Flussdiagramm eines Verfahrens zum Trainieren eines Algorithmus des maschinellen Lernens zum Ermitteln eines Ist-Zustandes von wenigstens einer Bauteileigenschaft eines wärmebehandelten metallischen Bauteiles gemäß Ausführungsformen der Erfindung;
• 2 ein Flussdiagramm eines Verfahrens zur Qualitätssicherung von wärmebehandelten metallischen Bauteilen gemäß Ausführungsformen der Erfindung;
• 3 ein Blockschaltbild eines Systems zur Qualitätssicherung von wärmebehandelten metallischen Bauteilen gemäß Ausführungsformen der Erfindung.

Show it:

• 1 a flowchart of a method for training a machine learning algorithm for determining an actual state of at least one component property of a heat-treated metallic component according to embodiments of the invention;
• 2 a flowchart of a method for quality assurance of heat-treated metallic components according to embodiments of the invention;
• 3 a block diagram of a system for quality assurance of heat-treated metallic components according to embodiments of the invention.

In the figures of the drawings, the same reference symbols designate the same or functionally identical elements, parts or components, unless otherwise stated.

1 1 shows a flowchart of a method 1 for training a machine learning algorithm for determining an actual state of at least one component property of a heat-treated metallic component according to embodiments of the invention.

Heat treatment methods influence, among other things, a surface zone of a metallic component, with the formation of the surface zone and its properties having a significant impact on the usage behavior and service life of the metallic component. Since conventional heat treatment processes usually do not allow any statements to be made about the final state of the surface zone of the heat-treated component, a final, precise examination of the heat-treated surface zone must usually be carried out after the heat treatment has taken place in order to make statements about the component properties such as hardening depth, hardness, microstructure, etc. to be able to meet, and thus about the usage behavior and service life of the corresponding heat-treated component.

It is known, for the final assessment of the heat-treated component and in particular the edge zone of the heat-treated component, to subject this to a visual check, which, however, has only a limited, subjective significance and is restricted to surface areas. For a further qualitative assessment of the heat-treated component, metallographic observations of the microstructure and hardness measurements in the section as well as residual stress and residual austenite measurements can be carried out using an X-ray diffractometer, although this sometimes requires the component to be destroyed. Test methods based on micromagnetic measuring methods also allow comprehensive and non-destructive testing of the surface zone properties of heat-treated components and the quantitative determination of various material parameters, but only provide locally valid information and are limited to magnetizable materials.

According to the embodiments of 1 the method 1 has a step 2 of acquiring data characterizing the heat treatment during the heat treatment of a metallic component, a step 3 of determining the actual state of the at least one component property from the acquired data, and a step 4 of training a Algorithm of machine learning based on the collected data and the, the current state of the at least one component property characterizing features on.

The machine learning algorithm for determining an actual state of at least one component property of a heat-treated metallic component can be used to achieve a quick, effective and complete inspection of the quality of heat-treated metallic components in a simple manner. The machine learning algorithm can be used to check the quality of heat-treated components in almost real time, especially immediately after the heat treatment has been carried out, without the need for complex testing methods following the heat treatment, such as a visual check, metallographic microstructure observations and hardness measurements in the section as well as Residual stress and residual austenite measurements using an X-ray diffractometer, or test methods based on micromagnetic measuring methods would be required. During quality inspection, poor-quality heat-treated components that may pose a safety risk can then be sorted out without having to exclude the entire batch of heat-treated components at the same time, since each component can be individually inspected based on the machine learning algorithm. Overall, based on the machine learning algorithm for determining an actual state of at least one component property of a heat-treated metallic component, improved testing of the quality of a heat-treated component is made possible.

The recorded data characterizing the heat treatment can be amplified and filtered before being digitized for further processing.

Step 3 of determining the actual state of the at least one component property from the recorded data can also include, for example, applying transformation methods to determine the Fourier spectrum, the wavelet spectrum and the Hilbert spectrum to examine and map the frequency range . The actual state of the at least one component property characterizing features are then derived in the form of statistical parameters such as effective values, optima or mean values from the generated transformations and recorded data in the corresponding time range. Overall, the data or raw data characterizing the heat treatment are thus mapped into a time-frequency range.

All common methods can be used to train the machine learning algorithm, and neural architecture search processes and/or hyperparameter optimization can also be used to optimize the machine learning algorithm. It is also possible to use only 80% of the training data effectively for training the machine learning algorithm, while the other 20% can be used for evaluating the trained machine learning algorithm.

The recorded data characterizing the heat treatment and thus training data are also sensor data.

According to the embodiments of 1 the sensor data include structure-borne noise signals, vibration signals and data about a current electrical output and a current current intensity of a corresponding heat treatment device.

In this case, the sampling rate of the structure-borne noise signal can preferably be 1 to 2 MHz and the sampling rate of the vibration signal and the electrical power and the current intensity of the heat treatment device can preferably be 50 to 100 kHz.

In addition, the method 1 according to the embodiments of 1 further comprises a step 5 of extracting non-explanatory features characterizing the actual state of the at least one component property prior to the training of the machine learning algorithm.

For example, a relative subset can be selected from existing features using filter, wrapper, or embedded approaches. Furthermore, feature extraction methods can also be used, in which case the overall information content of the data can be increased by combining existing features.

In addition, another machine learning algorithm can also be trained to monitor the components and resources of the heat treatment device itself.

2 10 shows a flow diagram of a method 10 for quality assurance of heat-treated metallic components according to embodiments of the invention.

how 2 shows, the method in turn has a step 11 of acquiring data characterizing the heat treatment during the heat treatment of a metallic component, a step 12 of determining the actual state of the at least one component property from the acquired data, and a step 13 of a Training a machine learning algorithm based on the collected data and the features characterizing the actual state of the at least one component property.

In addition, the method 10 according to the embodiments of FIG 2 a step 14 of heat treatment of a metallic component by a heat treatment device in order to achieve a desired target state of at least one component property, a step 15 of determining an actual state of the at least one component property using the trained machine learning algorithm to determine an actual state of at least one component property of a heat-treated metallic component, a step 16 of comparing the actual state of the at least one component property with the desired target state, and a step 17 of discarding the component if the actual state of the at least one component property changes by more than one permissible threshold value deviates from the target state.

According to the method 10 according to the embodiments of FIG 2 Based on the trained machine learning algorithm, a prediction is made about the probability of whether or not the specified, desired properties can be detected in a previously heat-treated component. In addition, the degree of deviation can be estimated and subsequently based on what is permissible, in particular based on standards and committees given permissible threshold value, whether the component can be reused.

According to the embodiments of 2 the method 10 also has a step 18 of adapting operating parameters of the heat treatment device based on a deviation of the actual state of the at least one component property from the target state such that the actual state of the at least one component property during subsequent heat treatments of metallic components the heat treatment device should match the target state again if the actual state of the at least one component property deviates from the target state by more than a permissible threshold value.

3 10 shows a block diagram of a system 20 for quality assurance of heat-treated metallic components according to embodiments of the invention.

how 3 shows, the system 20 has a heat treatment device 21 for heat treatment of metallic components and a control unit 22 for quality assurance of heat-treated metallic components.

According to the embodiments of 3 the system 20 further has a control unit 23 for training a machine learning algorithm for determining an actual state of at least one component property of a heat-treated metallic component and sensors 24 for acquiring data characterizing the heat treatment during the heat treatment of a metallic component by heat treatment device 21.

The control unit 23 for training a machine learning algorithm for determining an actual state of at least one component property of a heat-treated metallic component is designed to train the machine learning algorithm based on data recorded by the sensors during heat treatments by the heat treatment device 21 . Control unit 22 is also designed for quality assurance of heat-treated metallic components, the quality of the component heat-treated by heat-treatment device 21 based on the corresponding information provided by control unit 23 for training a machine learning algorithm for determining an actual state of at least one component property of a heat-treated metallic component To evaluate the machine learning algorithm trained on the component.

According to the embodiments of 3 the sensors 24 have a structure-borne noise 25 and a vibration sensor 26, which are mounted at the smallest possible distance from the contact zone between the tool and the workpiece or component.

Control unit 23 for training a machine learning algorithm for determining an actual state of at least one component property of a heat-treated metallic component also has a first receiving unit 27, which is designed to acquire data characterizing the heat treatment during the heat treatment of the metallic component, a first Determination unit 28, which is designed to determine the actual state of the at least one component property characterizing features from the recorded data, and a training unit 29, which is designed to use the machine learning algorithm based on the recorded data and the actual To train state of at least one component property characterizing features on.

The first receiving unit can also be a receiver, for example, which is designed to receive the data characterizing the heat treatment. The first determination unit and the training unit can also each be implemented, for example, based on corresponding code stored in a memory and executable by a processor.

In addition, the illustrated control unit 23 for training a machine learning algorithm for determining an actual state of at least one component property of a heat-treated metal component also has an extraction unit 30, which is designed to extract non-meaningful features characterizing the actual state of the at least one component property before the Training the machine learning algorithm to extract.

In this case, the extraction unit can, for example, in turn be implemented on the basis of appropriate code stored in a memory and executable by a processor.

According to the embodiments of 3 has the control unit 22 for quality assurance of heat-treated metallic components, a second receiving unit 31, which is designed, a through the control unit 23 for training an algorithm of machine learning to determine an actual state of at least one component property of a heat-treated metal machine learning algorithm for determining an actual state of at least one component property of a heat-treated metal component, a second determination unit 32, which is designed to receive an actual state of the at least one component property of a metal component heat-treated by a heat treatment device based on the machine learning algorithm for determining an actual state of at least one component property of a heat-treated metal component, a comparator 33, which is designed to compare the determined actual state of the at least one component property with a target state, and an output unit 34, which is designed to output that the metallic component should be discarded if the actual state of the at least one component property deviates from the target state by more than a permissible threshold value.

The second receiving unit can in turn be a receiver, for example, which is designed to receive the machine learning algorithm. The second determination unit, the comparator and the output unit can in turn each be implemented, for example, based on corresponding code stored in a memory and executable by a processor.

In addition, the illustrated control unit 22 for quality assurance of heat-treated metallic components also has an adjustment unit 35, which is designed to initiate, if the actual state of the at least one component property deviates from the target state by more than a permissible threshold value, that operating parameters of the heat treatment device 21 be adjusted based on a deviation of the actual state of the at least one component property from the target state in such a way that the actual state of the at least one component property should match the target state again during subsequent heat treatments of metallic components by the heat treatment device 21 .

The adjustment unit can be, for example, actuators correspondingly coupled to the heat treatment device.

QUOTES INCLUDED IN DESCRIPTION

This list of the documents cited by the applicant was generated automatically and is included solely for the better information of the reader. The list is not part of the German patent or utility model application. The DPMA assumes no liability for any errors or omissions.

Patent Literature Cited

• WO 2015169309 A1 [0005]

Claims (14)

Method (1) for training a machine learning algorithm for determining an actual state of at least one component property of a heat-treated metallic component, the method (1) having the following steps: - During the heat treatment of a metallic component, acquiring data (2) characterizing the heat treatment; - Determination of the actual state of the at least one component property characterizing features (4) from the recorded data (3); and - Training the machine learning algorithm based on the recorded data and the features (4) characterizing the actual state of the at least one component property. procedure after claim 1 , the recorded data (2) characterizing the heat treatment being sensor data. procedure after claim 2 , wherein the sensor data include structure-borne noise signals, vibration signals and data on a current electrical power, a current current and / or a currently applied voltage of a corresponding heat treatment device. Procedure according to one of Claims 1 until 3 , wherein the method (1) further comprises the following step: - Extraction of non-significant features (4) characterizing the actual state of the at least one component property before the training of the machine learning algorithm. Method for quality assurance of heat-treated metallic components, the method (10) having the following steps: - heat-treating a metallic component using a heat-treatment device in order to achieve a desired target state of at least one component property (14); - Determining an actual state of the at least one component property by one according to one of Claims 1 until 4 trained machine learning algorithm for determining an actual state of at least one component property of a heat-treated metallic component (15); - Comparing the actual state of the at least one component property with the desired target state (16); and - discarding the component if the actual state of the at least one component property deviates from the target state by more than a permissible threshold value (17). procedure after claim 5 , wherein the method (10) further comprises the following step: - If the actual state of the at least one component property deviates from the target state by more than a permissible threshold value, adjusting operating parameters of the heat treatment device based on a deviation of the actual state of the at least a component property from the target state such that the actual state of the at least one component property should match the target state again during subsequent heat treatments of metallic components by the heat treatment device (18). Control unit for training a machine learning algorithm for determining an actual state of at least one component property of a heat-treated metal component, the control unit (23) having a receiving unit (27) which is designed to transmit data characterizing the heat treatment during the heat treatment of a metal component detect, a determination unit (28), which is designed to determine the actual state of the at least one component property characterizing features from the recorded data, and a training unit (29), which is designed to use the machine learning algorithm based on the recorded data and to train the actual state of the at least one component property characterizing features. control unit claim 7 , where the data collected is sensor data. control unit claim 8 , wherein the sensor data include structure-borne noise signals, vibration signals and data on a current electrical power, a current current and / or a currently applied voltage of a corresponding heat treatment device. Control unit according to one of Claims 7 until 9 , wherein the control unit (23) further has an extraction unit (30), which is designed to extract non-meaningful features that characterize the actual state of the at least one component property before the machine learning algorithm is trained. System for training a machine learning algorithm for determining an actual state of at least one component property of a heat-treated metallic component, the system having at least one sensor (24,25,26) for detecting data characterizing a heat treatment of a metallic component during the heat treatment of the metallic component and a control unit (23) to Trai ning of a machine learning algorithm for determining an actual state of at least one component property of a heat-treated metallic component according to one of Claims 7 until 10 having. Control unit for quality assurance of heat-treated metal components, the control unit (22) having a receiving unit (31) which is designed to use a control unit to train a machine learning algorithm to determine an actual state of at least one component property of a heat-treated metal component one of the Claims 7 until 10 trained machine learning algorithm for determining an actual state of at least one component property of a heat-treated metal component, a determination unit (32) which is designed to determine an actual state of the at least one component property of a metal component heat-treated by a heat treatment device based on the machine learning algorithm for determining an actual state of at least one component property of a heat-treated metal component, a comparator (33) which is designed to compare the determined actual state of the at least one component property with a desired target state, and an output unit (34) which is designed to output that the metallic component should be discarded if the actual state of the at least one component property deviates from the target state by more than a permissible threshold value. System for quality assurance of heat-treated metallic components, the system (20) having a heat treatment device (21) for heat-treating metallic components and a control device (22) for quality assurance of heat-treated metallic components claim 12 having. system after Claim 13 , wherein the control unit (22) for quality assurance of heat-treated metallic components further an adjustment unit (35), which is designed to initiate if the actual state of the at least one component property deviates from the target state by more than a permissible threshold value, that operating parameters of the heat treatment device (21) based on a deviation of the actual state of the at least one component property from the target state in such a way that the actual state of the at least one component property during subsequent heat treatments of metallic components by the heat treatment device (21) again with the Target. State should match, has.

Images