Method for operating a beam device, computer program and beam device for carrying out the method as well as a method for generating a training data set and a method for training a machine learning model

A non-transitory computer readable medium containing software that is loadable into a processing unit and which, when executed by the processing unit, controls a beam device to perform the following:

• obtaining first data from a database that is connected to the processing unit, wherein the first data includes object data about the object, data about a plurality of machine learning models and first range data about a range of possible values of at least one first control parameter used to control a control unit of the beam device;
• determining a first machine learning model of the plurality of machine learning models based on the first data;
• providing the object data and the first range data as input data to the first machine learning model;
• determining, using the first machine learning model, a control value of the first control parameter, wherein the control value of the first control parameter is an output of the first machine learning model;
• obtaining second data from the database, wherein the second data includes the object data about the object, the data about the plurality of machine learning models as well as second range data about a range of possible values of at least one second control parameter to control the control unit of the beam device;
• determining a second machine learning model of the plurality of machine learning models based on the second data;
• providing the object data and the second range data as input data to the second machine learning model;
• determining, using the second machine learning model, a control value of the second control parameter, wherein the control value of the second control parameter is an output of the second machine learning model;
• using the control value of the first control parameter and the control value of the second control parameter to operate the control unit;
• and
• generating, using the beam device and the control unit of the beam device, the information about the object.

A beam device for generating information about an object, comprising:

• at least one beam generator that generates a beam;
• at least one objective lens that focuses the beam onto the object;
• at least one detector unit that detects interaction particles and/or interaction radiation resulting from an interaction of the beam with the object;
• at least one processing unit coupled to the beam device;
• and
• at least one non-transitory computer readable medium, coupled to the at least one processing unit and containing software that is loadable into the at least one processing unit and which, when executed by the at least one processing unit, controls the beam device to perform the following:
• obtaining first data from a database that is connected to the at least one processing unit, wherein the first data includes object data about the object, data about a plurality of machine learning models and first range data about a range of possible values of at least one first control parameter used to control a control unit of the beam device;
• determining a first machine learning model of the plurality of machine learning models based on the first data;
• providing the object data and the first range data as input data to the first machine learning model;
• determining, using the first machine learning model, a control value of the first control parameter, wherein the control value of the first control parameter is an output of the first machine learning model;
• obtaining second data from the database, wherein the second data includes the object data about the object, the data about the plurality of machine learning models as well as second range data about a range of possible values of at least one second control parameter to control the control unit of the beam device;
• determining a second machine learning model of the plurality of machine learning models based on the second data;
• providing the object data and the second range data as input data to the second machine learning model;
• determining, using the second machine learning model, a control value of the second control parameter, wherein the control value of the second control parameter is an output of the second machine learning model;
• using the control value of the first control parameter and the control value of the second control parameter to operate the control unit;
• and
• generating, using the beam device and the control unit of the beam device, the information about the object.

A method for generating a training dataset for a processing unit of a beam device and/or a machine learning model that determines a control value of a first control parameter to operate a control unit of the beam device and a control value of a second control parameter to operate the control unit of the beam device, the method comprising:

• determining the control value of the first control parameter by obtaining, using the processing unit, first data from a database that is connected to the processing unit, wherein the first data includes object data about the object, data about a plurality of machine learning models and first range data about a range of possible values of the first control parameter used to control a control unit of the beam device, determining, using the processing unit, a first machine learning model of the plurality of machine learning models based on the first data, providing the object data and the first range data as input data to the first machine learning model, and determining, using the first machine learning model, the control value of the first control parameter, wherein the control value of the first control parameter is an output of the first machine learning model;
• determining the control value of the second control parameter by obtaining, using the processing unit, second data from a database that is connected to the processing unit, wherein the second data includes object data about the object, data about a plurality of machine learning models and second range data about a range of possible values of the second control parameter used to control a control unit of the beam device, determining, using the processing unit, a second machine learning model of the plurality of machine learning models based on the second data, providing the object data and the second range data as input data to the second machine learning model, and determining, using the second machine learning model, a control value of the second control parameter, wherein the control value of the second control parameter is an output of the second machine learning model;
• generating, using the beam device and the control unit of the beam device, the information about the object;
• and
• storing the determined control value of the first control parameter, the control value of the second control parameter and the information about the object in a database as the training dataset.

A method for training a machine learning model the of a beam device to identify an image of an object which fulfils quality standards given by a user, the method comprising:

• a user of the beam device providing a first image of an object using a beam device, wherein the first image fulfils the quality standards;
• providing at least a first deteriorated image by deteriorating the first image using a processing unit associated with the beam device and providing at least a second deteriorated image by deteriorating the first image using the processing unit associated with the beam device;
• the user evaluating whether the at least one first deteriorated image fulfils the quality standards and labeling the at least one first deteriorated image depending on evaluating the at least one first deteriorated image;
• the user evaluating whether the at least one second deteriorated image fulfils the quality standards and labeling the at least one second deteriorated image depending on evaluating the at least one second deteriorated image;
• simulating, using the machine learning model, evaluating the at least one first deteriorated image and evaluating the at least one second deteriorated image, wherein the first image, the at least one first deteriorated image and the at least one second deteriorated image are used as training data for the machine learning model and wherein simulating is based on the first image, the at least one first deteriorated image and the at least one second deteriorated image;
• the user of the beam device providing a second image of the object using the beam device, wherein the second image fulfils the quality standards;
• providing at least a third deteriorated image by deteriorating the first image and/or the second image using the processing unit associated with the beam device and providing at least a fourth deteriorated image by deteriorating the first image and/or the second image using the processing unit associated with the beam device;
• evaluating, using the machine learning model, whether the at least one third deteriorated image fulfils the quality standards and labeling the at least one third deteriorated image depending on evaluating the at least one third deteriorated image, wherein evaluating and labeling the at least one third deteriorated image are repeated, until correctness of the label is verified, using the processing unit associated with the beam device;
• evaluating, using the machine learning model, whether the at least one fourth deteriorated image fulfils the quality standards and labeling the at least one fourth deteriorated image depending on evaluating the at least one fourth deteriorated image, wherein evaluating and labeling the at least one fourth deteriorated image are repeated, until correctness of the label is verified, using the processing unit associated with the beam device;
• the user and/or the processing unit associated with the beam device providing at least one third image of the object;
• and
• evaluating, using the machine learning model, whether the at least one third image fulfils the quality standards and labeling the at least one third image depending on evaluating the at least one third image, wherein evaluating and labeling the at least one third image are repeated, until correctness of the label is verified, using the processing unit associated with the beam device.