Vibration isolation system with diagnostic function and diagnostic procedures

Abstract

The invention is based on the objective of further improving or better assessing the performance of the vibration isolation system and thus the performance of the machine or device connected to the vibration isolation system. For this purpose, a vibration isolation system is provided by means of which a vibration-isolated load can be supported, comprising at least one vibration isolator, preferably a plurality of vibration isolators, which can counteract vibrations occurring during operation by means of actuators, and at least one sensor, preferably a plurality of sensors, which are configured to record measured values ​​relating to the vibration isolation system and / or the supported load and / or the environment and to provide these as input data to a computer-aided method for a performance analysis of the vibration isolation system and / or the supported load.

Description

[0001] The invention relates to a preferably stationary vibration isolation system with a diagnostic function, as well as a diagnostic method. In particular, the invention relates to an active, stationary vibration isolation system on which a machine for processing semiconductor components and / or nanostructured substrates or elements, in particular wafers, masks, or displays, such as flat-panel displays, or for measuring semiconductor components and / or nanostructured substrates, such as an electron beam microscope, is arranged.

[0002] Furthermore, laboratory equipment and / or medical equipment, such as imaging examination devices such as magnetic resonance imaging (MRI) scanners or microscopes, can also be arranged on the stationary vibration isolation system. Background of the invention

[0003] Active, stationary vibration isolation systems are used particularly in the semiconductor industry for the mounting of processing machines, such as lithography devices and measuring devices, such as electron beam microscopes, in order to provide vibration isolation for sensitive devices used in the processing of semiconductor components.

[0004] Such a vibration isolation system for a machine or machine parts usually consists of a plate which is mounted in a vibration-isolated manner on at least three, often four, and possibly more than four vibration isolators, each of which comprises a spring.

[0005] Alternatively or additionally, one or more insulators can be used that are not mounted on a plate, but are arranged between two parts within the machine, for example, between a base frame and a stage or a granite slab. Insulators can also be arranged outside the machine; for example, the entire machine can be placed on one or more insulators.

[0006] The machine or machine parts are thus protected from vibrations. The machine or machine parts, together with a plate, form the vibration-isolated load.

[0007] Active vibration isolation systems are particularly well known in practice. These include spring-loaded vibration isolators and actuators, particularly magnetic actuators, that actively counteract vibrations. Vibration isolation systems also exist in which additional sensors or actuators are mounted at other locations on the machine, separate from the vibration isolators. Vibrations are detected via sensors, which can be located at the base of the vibration isolator and / or near the base of one or more vibration isolators, as well as on the vibration-isolated load. Based on the sensor signals, the actuators are controlled to compensate for vibrations by generating counterforces.

[0008] By means of such an active control, both vibrations that are prolonged from outside into the system and vibrations that are generated by the vibration-isolated load itself (e.g. from a movable platform, such as a stage, or a moving robot) can be counteracted.

[0009] The applicant's patent EP 2 295 829 B1 shows a vibration isolation system in which actuators for generating compensation forces are present in several degrees of freedom.

[0010] Such vibration isolation systems comprise a control unit (often referred to as a controller) for processing the sensor signals. This control unit records the sensor signals and, based on the sensor signals, generates control signals by means of which the actuators are controlled in order to generate compensating forces.

[0011] The control unit can be designed as a central processing unit to which both the sensors and actuators of all isolators of the vibration isolation system are connected. The applicant's document EP 3 726 094 A1 describes a method for controlling such a stationary vibration isolation system.

[0012] The invention is based on the object of further improving or better assessing the performance of the vibration isolation system and thus the performance of the machine or device connected to the vibration isolation system, or of optimizing the availability of the machine.

[0013] In particular, the invention is based on the object of obtaining further data on the vibration isolation system and / or on the vibration-isolated machines or devices, which can be used, for example, to determine or predict the performance of the machines or devices, or which allow changes on or in the environment of the machine, such as signs of wear, to be detected in a timely manner and provide indications for remedial action. In this context, the performance of the machines refers to their precision, accuracy, availability, and productivity.

[0014] The inventors have taken on this task.

[0015] This task is achieved by a vibration isolation system with diagnostic function and a method for diagnosing a vibration isolation system and / or a load mounted thereon in a vibration isolation manner according to one of the independent claims. Preferred embodiments and further developments of the invention can be found in the respective subclaims.

[0016] The subject of the invention is therefore, in a first aspect, a Vibration isolation system, in particular with diagnostic function, preferably stationary, active vibration isolation system, by means of which a vibration-isolated load can be supported, comprising at least one vibration isolator, preferably a plurality of vibration isolators, which can counteract vibrations occurring during operation by means of actuators, at least one sensor, preferably a plurality of sensors, which are configured to record measured values ​​relating to the vibration isolation system and / or the supported load and / or the environment and to provide them as input data to a computer-aided method for a performance analysis of the vibration isolation system and / or the supported load.

[0017] The vibration isolators can actively or passively counteract the vibrations that occur. Active and passive vibration isolators can also be combined to form a vibration isolation system.

[0018] Measurement values ​​relating to the supported load can, in particular, relate to a machine or parts thereof mounted with the vibration isolation system. For example, a vibration sensor can be attached to an optical sensor of such a machine, or in the case of an electron microscope, to the electron beam column. The sensor itself does not even have to be attached to the supported load or the vibration isolation system. Thus, vibrations of the system, the load, or parts thereof could also be recorded using optical sensors located outside of this system, for example.

[0019] For the purposes of this disclosure, a performance analysis is understood to mean a classification of the operating state with regard to harmful disturbances and influences. In the case of a vibration isolation system, these disturbances particularly represent influences due to undesired movements of the system, especially vibratory movements. In addition to the classification of disturbance variables, the performance analysis also includes the classification of whether the system is operating well, in particular optimally, including whether the system is optimally adjusted or set. For example, performance can be classified as good or poor with regard to the decay time after the movement of a component of the system, such as after the movement of an xy table, but also the resulting frequency responses or other environmental conditions, such as other processes or operating states running simultaneously in the machine, temperatures, load cases on the stage.

[0020] In general, with regard to performance analysis, a distinction can be made between operating states and processes or machine functions. In both cases, forces from the machine's actuators or from the environment are introduced into the dynamic system, the structural dynamics of the machine. In structural dynamics, vibrations are excited, which can be recorded by the sensors of the vibration isolation system. In operating states, quasi-static force introduction or constant vibration states are assumed. In processes or machine functions, transient force introductions or vibration states occur. In both cases, the sensors of the vibration isolation system can record and evaluate the amplitude and spectral composition of the vibration signals.For processes or machine functions, the temporal behavior of the vibration amplitudes and the spectral composition are also taken into account. These observations can then be used as a performance analysis to derive evaluations and classifications for the respective operating states, processes, or machine functions.

[0021] Performance analysis therefore specifically involves evaluating a multitude of measured values ​​provided by the sensor(s) of the vibration isolation system, as well as assessing the system's operating status, expressed as an output value. However, additional or different information can also be used or utilized, such as whether the machine is producing the required quality. The output value can comprise one of several classes that characterize the system's operating status. In the simplest case, a classification is made into just two classes, which characterize a good and a poor operating status. The output value can also be a sliding value within a predefined range, for example, in the form of a percentage. For example, an output value of 100% could indicate an optimal operating status, while a value of 75% could indicate a just-acceptable status.

[0022] Of particular importance for the performance of a vibration isolation system are naturally movements, especially vibrations and their suppression.Therefore, according to a second aspect of this disclosure, a vibration isolation system for the vibration-isolated mounting of a load is provided, comprising at least one, preferably a plurality of vibration isolators and at least one sensor, wherein the sensor is configured to detect a movement variable, in particular a mechanical vibration, movement, or acceleration, and a computing device, wherein the computing device contains an artificial neural network, wherein the artificial neural network is configured to process data of the movement variable provided by the at least one sensor and to classify whether the vibration isolation system with the mounted load is in an operating state with sufficiently little or excessive movement disturbance. This classification can then be output.In a simple case, the output can be intended to inform operating or monitoring personnel about the operating status. The output can also be used by the vibration isolation system itself or another IT-connected device to influence the operating status if it has been classified as being disturbed by excessive movement. Motion variables that serve as input data can primarily be vibrations, in particular their amplitude and frequency. Other alternative or additional motion variables can also be accelerations and velocities. For the sake of simplicity, the aforementioned operating states are also referred to below as "good" (operating status disturbed sufficiently little by movement) or "poor" (operating status disturbed excessively by movement).Of course, as explained above, the classification can include further or finer gradations, for example in the form of a sliding output value.

[0023] A particularly preferred form of data output by the sensor and processed by the artificial neural network is spectrograms. For the purposes of this disclosure, a spectrogram is understood to be, in particular, a data set containing the temporal progression of the amplitudes of several oscillation frequencies.

[0024] The present invention makes it possible to use sensor readings as input data to obtain performance data of the vibration isolation system and / or the supported load.

[0025] The supported load can, for example, comprise processing machines or parts thereof, particularly in the semiconductor industry, such as machines for processing semiconductor components and / or nanostructured substrates or elements, particularly wafers, masks, or displays; lithography equipment; or measuring devices or equipment, particularly for measuring semiconductor components and / or nanostructured substrates, such as an electron beam microscope; or other sensitive equipment, machines, or systems, such as laboratory equipment or medical equipment, such as imaging examination equipment such as magnetic resonance imaging (MRI) scanners. For the sake of summary and simplicity, the terms "machine" and "device" will also be referred to generally below.

[0026] In many embodiments of the invention, the supported load can comprise a plate, particularly as a mount, on which the machine or device can be arranged and secured. Devices or machines arranged on such a plate are thus protected from vibrations. The plate and the devices or machines arranged thereon can together form the vibration-isolated load. The supported load can be supported by means of a three-point support via three, but also via four or more vibration isolators.

[0027] According to a preferred embodiment of the invention, the sensors are configured to record a variety of information or measured values ​​relating to the vibration isolation system and / or the supported load or the environment during operation and to make them available to suitable computer-aided methods. According to the invention, this input data can be used for a performance analysis of the vibration isolation system and / or the supported load. The result of the performance analysis can be used to diagnose the vibration isolation system and / or the supported load, so that the vibration isolation system can have a diagnostic function.

[0028] According to a preferred embodiment of the invention, the sensors can provide a wide range of information, for example, measured values ​​that can be viewed in specific contexts, and whose evaluation using rule-based or statistical methods is only of limited use or even impossible. This can be of great interest, for example, when the vibration isolation system is used on or with machines or systems in the semiconductor industry, in the field of inspection or measurement technology, or even generally for use on or with machines.

[0029] For example, with certain types of machines, it can be helpful to reduce or eliminate vibrations as much as possible. The performance, such as the optical resolution of a measuring or inspection system, can depend heavily on the vibrations that occur.

[0030] Accordingly, there can be a diverse relationship between the vibrations occurring and the performance of the load supported by the vibration isolation system.

[0031] To reduce vibrations, sensors can be provided that can measure vibrations occurring at defined points in the system. These can be, for example, acceleration sensors that can detect acceleration in one of the spatial directions x, y, or z of a Cartesian coordinate system. Since in many cases there can be a correlation between the performance of the vibration isolation system and the supported load (in particular, the performance of the supported machine or device) and the vibrations that occur, and the vibrations can also be measured at the same time, it is possible to make a statement about the performance of the supported load, or rather the supported machine or device, based on the measured values ​​of the sensors.

[0032] The inventors have discovered that even more information can be obtained from the sensor measurements. For example, it can be determined whether a machine on bearings is still operating within its specifications. Based on the vibration characteristics, the cause of a possible deviation from the specification can also be determined, for example, whether a bearing-mounted xy table or a table with a bearing is defective.

[0033] The cross table ("stage") is working correctly, or whether, for example, a vacuum pump connected to the system is transmitting too many vibrations to the system, or even whether there is a seismic cause for the deviation. For the purposes of this disclosure, the characteristics of the vibrations are understood to mean temporal changes in the spectral distribution and correlations between signals, with one signal representing the excitation and the other signal(s) representing a response. In general, the characteristics of the vibrations can also be represented by the time-dependent intensity of the various frequencies of the occurring frequency spectrum, for example, after a pulse excitation.

[0034] Furthermore, information about the operating status of the machine can be obtained based on the characteristics of vibrations, for example, whether an xy table is in a specific position, such as a transfer position for a substrate, or whether it is still moving. Thus, the vibration spectrum can change depending on the position of the table. Certain vibration patterns can also indicate specific operating conditions or errors, such as a defective vacuum pump.

[0035] Furthermore, the vibration characteristics can provide information about wear and tear or the maintenance status of the machine. This information can also be made available to the machine control software or the operating personnel.

[0036] The current operating states of the stored load can also be taken into account, for example rest state, acceleration or movement of an xy table.

[0037] Other measurable factors influencing the performance of the machine include, for example, vibrations in the environment, such as the floor, especially at the installation site, or existing electromagnetic fields, or the temperature.

[0038] The object of the present invention is therefore to use information obtained by means of the sensors of a vibration isolation system not only directly for vibration isolation, i.e. the control of the actuators for vibration reduction, but also for further analyses of the performance of the vibration isolation system and / or the load mounted on it and in this way to carry out a performance analysis which enables a diagnosis.

[0039] According to one embodiment of the invention, it is also provided to use this information, for example, for the control and regulation of the vibration isolation system and / or the load mounted thereon, or the machine or device, or other system components such as a vacuum pump.

[0040] The evaluation of the measured values ​​can become very complex. Vibration data, for example, can involve different frequencies with different amplitudes, which can be represented in the time domain, the frequency domain, or as a spectrum, among other things. Additional environmental conditions or machine states that must be taken into account further increase the complexity.

[0041] Rule-based, computer-aided procedures and methods can be used for evaluation or performance analysis. Specific rules can be defined that can generate output information based on the measured values ​​or data. An example of a simple rule is that a machine is no longer operating within its specifications if the amplitude of a vibration exceeds a specified limit.

[0042] Therefore, analytical, statistical, or other mathematical methods can be used to generate the desired information according to defined analytical or statistical rules. These rules are generally generated manually.

[0043] This can provide information for a simple performance analysis, which can include, for example, information on the following aspects: - Earthquake or floor vibration, - “Windowing function”: is the xy table at the position for a transfer, - Monitoring of vibration levels according to limit values.

[0044] Statistical or other rule-based methods, which require the manual creation and storage of rules according to which the desired information is generated from the input data, reach their limits when the input data becomes too complex or interdependent.

[0045] For example, it may happen that a vibration or acceleration is permitted to exceed a limit if it occurs in a specific operating state of a machine as a suspended load, where this short-term exceedance is not or hardly relevant to performance, and / or if there are simultaneously compensating influences on the machine, and / or if the exceedance does not occur frequently. Manually defining rules for such complex relationships is not only very time-consuming but also highly error-prone, or even impossible or at least not reliably possible in complex relationships.

[0046] The invention therefore provides a computer-aided performance analysis based on artificial neural networks, which can be trained instead of fixed rules. This makes it possible to obtain significantly more information about the performance of the vibration isolation system and / or the supported load and make it available for control and regulation.

[0047] In one aspect of the invention, a vibration isolation system is provided which uses artificial intelligence methods or methods, preferably artificial neural networks, for performance analysis of the vibration isolation system and / or the supported load.

[0048] The input data obtained from the measured values ​​can relate to the vibration isolation system itself, the supported load, or other information. This input data can be used individually or in combination. For diagnostic purposes, the information can be classified, both individually and in combination.

[0049] Suitable sensors can be selected and made available to capture the desired measurement values. Any specific sensors mentioned below represent only a selection and should not be considered exhaustive.

[0050] The input data can, for example, relate to the vibration behavior of the vibration isolation system, including the signal distribution in the frequency spectrum – changes in the frequency domain – as well as the impulse and decay behavior – changes in the time domain. Suitable sensors can, for example, include acceleration sensors, preferably for measuring in three spatial directions.

[0051] To capture the underlying measured values, the sensors can be provided in or on the vibration isolation system and, for example, measure the vibration behavior in one or more directions. Sensors with one degree of freedom are preferably used. By using multiple such sensors, correspondingly multiple degrees of freedom, for example up to six, or vibration directions, can be captured. According to another embodiment, however, one or more sensors can also be used that can measure in more than one direction or in more than one degree of freedom.

[0052] For this purpose, the sensors can be arranged not only in or on the vibration isolation system, but also on the supported load, for example, on or in a supported machine or on or in a supported device. If the supported load comprises a plate, it may also be advantageous, according to a preferred embodiment of the invention, to assign the sensors to this plate.

[0053] Furthermore, the sensors can also be arranged on or in other associated components, for example on an xy table, on an optical element, on a casing, on a vacuum pump, or on another component of the vibration isolation system.

[0054] The input data can also include operating states ("on," "off," "at rest," "in process," "service," "fault") of the vibration isolation system and / or individual components of the supported load. In this way, information can be obtained regarding the standstill or operation of the vibration isolation system and / or the supported load, possible acceleration, the movement of an xy table, or the operating states of a component, such as a vacuum pump. Optical sensors, ultrasonic sensors, or distance sensors can be provided for this purpose.

[0055] Additional sensors can be assigned to the environment of the vibration isolation system, for example, the floor or the installation location, or to the surroundings, such as other components in the vicinity of the vibration isolation system. It is particularly advisable to select components that can transmit vibrations to the vibration isolation system and / or to the supported load.

[0056] Furthermore, it can be helpful to record the strength and spectral composition of electromagnetic fields prevailing at the installation location of the vibration isolation system, which can also impair the performance of the vibration isolation system and / or the supported load. Within the meaning of the invention, information other than vibration can also be used to determine machine performance data. For example, instead of a spectrum of a magnetic field, it can also be determined whether the machine is still operating within the specified specifications, i.e., within the permitted tolerances, with this magnetic field. A combination of a magnetic field and vibration data can also be used for this purpose. This is becoming increasingly complex, but this is precisely the advantage of the device described here, whose strength lies in processing complex relationships. In many cases, the performance of the supported load depends on various environmental variables - e.g.Magnetic field - dependent. By detecting various signals, the effect of combinations of different interference influences on the machine's performance can also be taken into account. Magnetic field sensors, fluxgate magnetometers, or saturation core magnetometers can be used to measure electromagnetic fields.

[0057] Furthermore, it may be advantageous to measure the strength of acoustic fields present at the installation site. Similar advantages apply here as with the previously described embodiments involving the detection of magnetic fields. The performance of the supported load often depends on various environmental variables – e.g., sound pressure. By detecting various signals, the effect of combinations of different interference influences on the machine's performance can also be taken into account. Pressure sensors and microphones, for example, can be provided for this purpose.

[0058] Furthermore, for the same reasons, it may also be advantageous to obtain information about the temperature at the installation site, and / or on components of the vibration isolation system, and / or the supported load. Appropriate thermocouples can be provided for this purpose.

[0059] In many cases, the performance of the supported load depends on various environmental variables, such as temperature. By recording various signals, the effect of combinations of different disturbances on the machine's performance can also be taken into account. This can make it possible to detect faulty operating conditions of the vibration isolation system or the supported load, for example, overheating of a faulty component, or to detect operating conditions where normal operation is no longer possible, for example, with optical measurements.

[0060] Based on the measured variables described above, it is clear that any correlations of any input variables with respect to the machine's performance data can be determined, provided they exist and are measurable. Without being limited to specific measured variables and their sensors, a further development of the invention therefore provides that the vibration isolation system is configured to classify the performance status of a device mounted with the vibration isolation system based on a correlation of the input data from the multiple sensors, which detect different measured variables.

[0061] According to the invention, a performance analysis can be carried out on the basis of the input data by means of the computer-aided methods, in particular with the support of artificial neural networks, in order to classify the performance of the vibration isolation system and / or the supported load.

[0062] The classification can, for example, include the characteristics "good" or "poor". Based on this performance analysis, the diagnosis can, for example, provide the evaluation "operation possible" or "operation not possible". A classification as "good" or "poor" can, for example, also refer to the throughput of a device mounted with the vibration isolation system. If the decay times after a device movement are too long and the process carried out by the device depends on whether the device is at rest, an unfavorable vibration spectrum leads to hold times and thus to a slowdown in production or measurement. If, on the other hand, the process does not take vibrations directly into account, the vibrations can in turn influence the measurement accuracy or production accuracy.In this regard, the condition of the vibration isolation system can also be classified as "good," "acceptable," or "poor." Further diagnoses could include "operation possible with restrictions" or "operation possible with changes to the control or regulation of a component."

[0063] The diagnosis can therefore, for example, concern the setting or adjustment of the vibration isolation system itself, whereby the vibration isolation system's specifications can be used as a reference value. A "good" classification can mean that the vibration isolation system is operating within specifications ("operation possible").

[0064] In a further embodiment of the invention, the supported load can be analyzed, for example, to determine whether a supported machine or device is operating or capable of operating within its specifications ("good") or not ("poor"). Here, too, the specifications of the machine or device can be used as a classification criterion. All of the signals described above can be used as input information as classification criteria, for example, the environmental specifications, such as floor vibration levels, or electromagnetic or acoustic environmental conditions and other vibration data. The various data can also be combined for classification purposes.

[0065] According to the invention, computer-assisted methods, preferably including or based on artificial neural networks, can be used to generate the output data from the input data. Suitable data preparation or processing methods, which will not be discussed in detail here, can be used to prepare the measured values ​​or generate the input data.

[0066] Different configurations of neural networks for machine learning can be used for the performance analysis according to the invention.

[0067] In a preferred embodiment of the invention, the artificial neural network can form at least one model comprising at least one convolutional layer and / or at least one neural layer. It can preferably further comprise at least one of the following additional components: an activation layer, a pooling layer, a flattening layer, and / or a drop layer. The artificial neural network advantageously comprises data for weighting at least one, preferably several, and particularly preferably all weightable components of the model. This weighting can be determined particularly easily through a training process.

[0068] Possible layers of a suitable artificial neural network will be briefly outlined here: - Input Layer: here the spectrogram is read into the network - Convolutional layer(s): A variety of configurations are possible (e.g. regarding the number of layers, the number and size of the filters, the activation functions, MaxPooling functions), - Flatten Layer: Converting the 3-dimensional result of the last convolutional layer into a one-dimensional format for transfer to the neural layers - Neural layer: A variety of configurations are possible (e.g., regarding the number of layers, the number of neurons, the activation functions, dropouts). In certain cases, it is also possible to omit the neural layers entirely (this also applies to the classification layer). In such a case, the classification can also be generated by a convolution layer. - Neural layer for classification: the neurons represent the desired classes, e.g., two neurons for the classes "good" and "bad." During the learning process, the network parameters are optimized through iterative target-actual comparison of the generated classification; these parameters are then applied during the classification process. A softmax activation can be used to generate a float value to represent the spectrogram's class membership in the two classes "good" and "bad."

[0069] The learning process of such artificial neural networks is often referred to as a self-adaptive algorithm or "machine learning." When a series of hierarchical levels or "layers" are used, this is also referred to as "deep learning," although the required number of "layers" is not precisely defined.

[0070] Using the example of input data concerning the vibration behavior of the vibration isolation system, the training or learning of the artificial neural network will be briefly outlined.

[0071] The measured value is the vibration frequency of the vibration isolation system, which is recorded using suitable sensors and represents the input for the artificial neural network as spectrograms. Each spectrogram can represent a process that can be classified as "good" or "bad," provided these two classes are to be formed for analysis. The next step is "labeling," i.e., assigning the spectrograms to the "good" or "bad" classes.

[0072] The generation of variances can be achieved by augmenting the spectrograms, i.e. by changing or transforming the data sets, i.e. deliberate deviations of the spectrograms or input variables are generated - preferably also with computer support - in order to increase the number of different training data sets.

[0073] In this way, i.e., by recording a large number of positioning sequences and possibly augmenting them, a large number of spectrograms can be generated, which can be used to train the artificial neural network used. The spectrograms are assigned to the designated categories or classes. According to one embodiment of the invention, the assignment can also be performed automatically with computer support.

[0074] For good performance analysis and diagnostics, it is helpful to have enough spectrograms available in the relevant classes. In certain cases, 10 to 100 input data sets, such as spectrograms, per class may be sufficient; however, approximately 1,000 or even 10,000 or more data sets per class are better to improve and increase the quality of the performance analysis.

[0075] The artificial neural network is created, for example, a convolutional network as described above. In the example discussed, the artificial neural network can have two output neurons, which represent the two defined classes "good" and "bad." This allows learning to occur. The training of an artificial neural network is generally known to those skilled in the art and will therefore not be discussed further here. The result is the so-called "weights file," which contains the parameterization of the individual elements and connections of the CNN network.

[0076] It is understood that the training procedure described above is only an example and that both the number and type of classes as well as the input variables can be different and that other architectures of the artificial neural networks as well as other input data can be used. For example, 4-dimensional input information can also be used, in which the fourth dimension can be a time axis, along which the times for different spectrograms lie.

[0077] It is further understood that in regular operation of the vibration isolation system, an already trained or learned artificial neural network is preferably used.

[0078] In an advantageous embodiment of the invention, the artificial neural network can be trained or further trained during regular operation of the vibration isolation system. This offers the advantage that the vibration isolation system according to the invention can be easily adapted to the conditions prevailing at the installation site.

[0079] In a further advantageous embodiment of the invention, it is provided that the artificial neural network uses learned data from a previous learning process or from another external source.

[0080] The performance analysis can generate information or output data, for example, control parameters for controlling or regulating the vibration isolation system and / or the supported load. An important aspect of the invention is therefore that these control parameters can be used directly or indirectly to control or regulate the vibration isolation system and / or the supported load.

[0081] The invention therefore also encompasses a method for controlling or regulating a vibration isolation system, wherein, during operation, control parameters for controlling the vibration isolation system are obtained and used as part of a performance analysis as described above. In this way, a self-learning vibration isolation system can be provided that can react, for example, to changes in the environment, such as vibrations occurring via the floor.

[0082] The vibration isolation system may comprise an active, stationary vibration isolation system as described above. A load mounted in a vibration-isolated manner may be supported by a plurality of vibration isolators, and vibrations may be actively counteracted by a plurality of actuators.

[0083] In a further aspect, the invention also includes a method for diagnosing a vibration isolation system, comprising a vibration isolation system as described above.

[0084] Also encompassed by the invention in yet another aspect is a processing machine, in particular in the semiconductor industry, such as a machine for processing semiconductor components and / or nanostructured substrates or elements, in particular wafers, masks or displays, a lithography device, or a measuring device or device, in particular for measuring semiconductor components and / or nanostructured substrates, such as an electron beam microscope, or another sensitive device, a machine or system, for example a laboratory device or a medical device, for example an imaging examination device such as a magnetic resonance imaging (MRI) scanner, comprising a vibration isolation system as described above.

[0085] The method can enable an analysis of the performance of the vibration isolation system, but also of the supported load.

[0086] Further details of the invention emerge from the description of the illustrated embodiments and the appended claims. The drawings show: Fig. 1 shows a schematic plan view of the basic structure of a stationary vibration isolation system, Fig. 2 schematically shows an example of the structure of the use of an artificial neural network with machine learning for the evaluation of input data, Fig. 3 schematically shows a possible structure of a suitable artificial neural network, Fig. 4 schematically shows a process for training an artificial neural network, Fig. 5 the vibration behavior of an xy-table in the three directions of a Cartesian coordinate system during different operating phases, Fig. 6 shows an example of a spectrogram, Fig. 7 shows, by way of example, the process of training an artificial neural network using spectrograms according to a first embodiment, Fig. 8 shows an example of the classification process based on spectrograms obtained during regular operation according to the embodiment of Fig. 7, Fig. 9 shows an example of a spectrogram according to a second embodiment, Fig. 10 shows, by way of example, the process of training an artificial neural network using spectrograms according to the second embodiment, Fig. 11 shows an example of the classification process based on spectrograms obtained during regular operation according to the embodiment of Fig. 10, Fig. 12 shows exemplary input data according to a third embodiment, Fig. 13 shows, by way of example, the process of training an artificial neural network based on input data according to the third embodiment, Fig. 14 shows an example of the classification process according to the embodiment from Fig. 13, Fig. 15 shows exemplary input data according to a fourth embodiment, Fig. 16 shows, by way of example, the process of training an artificial neural network based on input data according to the third embodiment, Fig. 17 shows an example of the classification process according to the embodiment from Fig. 15, and Fig. 18 schematically shows a process for adapting an artificial neural network. Detailed description of preferred embodiments

[0087] In the following detailed description of preferred embodiments, for the sake of clarity, like reference numerals designate substantially similar parts in or on these embodiments. However, to better illustrate the invention, the preferred embodiments illustrated in the figures are not always drawn to scale.

[0088] Fig. 1 shows a schematic plan view of the basic structure of a vibration isolation system 1. The vibration isolation system 1, which is also shown schematically, comprises a plurality of vibration isolators 2a - 2d, which comprise a spring and support a vibration-isolated load 4. The supported load 4 can be supported by means of a three-point support via three, but also via four vibration isolators 2a-2d, as in the example of Fig. 1 shown, stored.

[0089] In the example shown, the supported load 4 comprises only the plate; a vibration-isolated machine or device is not shown for clarity. In other words, during normal operation, the supported load typically includes the plate and the vibration-isolated machine or device.

[0090] The stored load 4 can, for example, comprise processing machines or parts of such machines, in particular in the semiconductor industry, such as machines for processing semiconductor components and / or nanostructured substrates or elements, in particular wafers, masks or displays, lithography devices, or measuring devices or devices, in particular for measuring semiconductor components and / or nanostructured substrates, such as an electron beam microscope, or other sensitive devices, machines or systems, for example laboratory devices or medical devices, such as imaging examination devices such as magnetic resonance imaging (MRI).

[0091] Each of the vibration isolators 2a-2d comprises actuators 5a-5d and sensors 6a-6d. The sensors 6a-6d detect vibrations of the vibration-isolated load 4 and / or the base of the vibration isolator coupled to the floor. The sensors 6a-6d, like the actuators 5a-5d, are connected to a computing device 3, which is centrally located in the example. Based on the signals from the sensors 6a-6d, the computing device 3 calculates compensation signals for controlling the actuators 5a-5d. In this case, the computing device 3 also functions as a control unit.

[0092] The vibration isolators 2a-2d are preferably effective in both the vertical and horizontal directions, as in the exemplary embodiment. The vibration isolators 2a-2d can comprise a spring, in particular a pneumatic spring, for this purpose. Sensors, spring, and actuators can be integrated components of such a vibration isolator or can be designed as separate units. In the latter case, the separate, particularly spatially adjacent units of sensors, spring, and actuators together form a logical unit that functions as an active vibration isolator, which not only acts as a passive spring but actively counteracts vibrations.

[0093] The vibration isolation system 1 according to the invention comprises a plurality of actuators, by means of which vibrations are actively counteracted by generating forces acting on the vibration-isolated load. The actuators can be designed, in particular, as magnetic actuators, as in the exemplary embodiment, or as pneumatic actuators. Control parameters for controlling or regulating the vibration isolation system 1 can be converted in the form of control signals via a digital-to-analog converter (not shown), for example, into a current to drive the magnetic actuators.

[0094] The computing device 3 is configured to execute a program by which the control signals are generated, taking into account the measured values ​​of the sensors, in order to generate the compensating forces. The control signals are signals that serve to control the actuators based on the sensor signals and to generate counterforces. Compensating forces are preferably generated in at least two, particularly preferably at least three, degrees of freedom via the actuators in the isolator.

[0095] The Fig. According to a preferred embodiment, the vibration isolation system 1 shown in Fig. 1 comprises actuators 5a-5d, via which compensation forces can be generated in three translational degrees of freedom according to the three spatial directions x, y and z and / or in three rotational degrees of freedom.

[0096] The vibration isolation system 1 according to the invention can be designed as a stationary, active vibration isolation system, meaning that during operation it is firmly positioned on a designated mounting surface and can actively counteract any vibrations that may be introduced, for example, via the floor. In another embodiment, the vibration isolation system is designed passively. In this case, the computing device is particularly designed to monitor the operating status of the system with the supported load based on the input data from the sensor(s) using the performance analysis.

[0097] The measured values ​​of the sensors are provided as input data to the computer-aided process, which is stored as a program in the computing device 3, and are used not only for vibration damping but also for a performance analysis of the vibration isolation system 1 and / or the supported load 4. Thus, performance data of the vibration isolation system 1 or the supported load 4 can be obtained.

[0098] The sensors 5a-5d are configured to record a variety of information or measured values ​​relating to the vibration isolation system 1 and / or the supported load 4 or the environment, for example, the installation surface, during operation, in order to enable a performance analysis of the vibration isolation system 1 and / or the supported load 4 as part of a diagnosis. For the sake of simplicity, the illustrated embodiment only shows the sensors 5a-5d associated with the plate. Additional sensors can be associated with a supported device or machine and / or the environment, for example, the floor or the installation surface, the wall, the ceiling, or similar.

[0099] According to the invention, a computer-aided performance analysis based on artificial neural networks is provided, which can be trained instead of fixed rules. Fig. 1, an artificial neural network 10 is schematically shown as a component of the computing device 3, in particular designed as a control unit, for illustrative purposes only.

[0100] The invention accordingly provides a vibration isolation system 1 which uses artificial intelligence methods, in particular artificial neural networks, for performance analysis and diagnosis of the vibration isolation system 1 and / or the supported load 4.

[0101] The input data obtained from the sensor measurements can relate to the vibration isolation system 1 itself, the supported load 4, or other information. This input data can be used individually or in combination. For diagnostic purposes, the information can be classified, both individually and in combination.

[0102] In the embodiment of the Fig. 1, the input data includes the vibration behavior of the plate. This can include the amplitude, period, frequency, and / or frequency spectrum of the plate. For this purpose, sensors 5a-5d are designed as acceleration sensors. In one example, the acceleration data in the x-direction of the acceleration sensor 6a of the isolator 2a is used.

[0103] The following describes an exemplary embodiment of the vibration isolation system 1 according to the invention, which is designed to enable the quality of the adjustment or adjustment of the vibration isolation system 1 into the classes "good" and "poor." It is understood that these two classes are merely examples, and other classes, such as "medium," or other classifications are also possible and contemplated. For the exemplary embodiment, a plate with an xy stage is assumed to be the vibration-isolated load 4. This xy stage enables movement in the spatial directions x and y.

[0104] It will be apparent to the person skilled in the art that the xy table specified in the embodiments is selected only as an example for illustration purposes and, of course, the underlying principles and methods also apply to other mounted machines and devices.

[0105] Fig. Figure 2 schematically illustrates the use of an artificial neural network with machine learning for evaluating input data using an example. From the input data, for example, the amplitude of a vibration over time or an acceleration representing the vibration, a spectrogram can be generated using data processing. In this example, this is the temporal progression of the frequency spectrum, which can be input into the artificial neural network. The output data is a classification, i.e., an assignment of the spectrogram to the classes "good" or "bad."

[0106] Fig. Figure 3 schematically shows a possible structure of a suitable artificial neural network. The numbers indicate the different levels or layers, which will only be briefly mentioned below for the sake of simplicity. (1) Input Layer (2) Convolutional layer(s) (3) Flatten Layer (4) Neural layer (5) Neural layer for classification.

[0107] In a suitable embodiment of the invention, the artificial neural network forms at least one model which comprises at least one convolutional layer and / or at least one neural layer.

[0108] The artificial neural network is fed with data, in this example, spectrograms, to weight at least one, preferably several, and especially preferably all weightable components of the model. This weighting can be determined particularly easily through a training process.

[0109] Fig. Figure 4 schematically shows a suitable, possible process for training an artificial neural network, wherein the initial data comprises two classes. The process comprises the steps designated in the flowchart with the reference numerals 101 to 109. In step 101, data acquisition is carried out. Step 102 is data processing. In step 103, labeling of the processed data is carried out. Subsequently, in step 104, augmentation takes place so that a set of training spectrograms (step 105) and a set of test spectrograms (step 106) are obtained. This creates a network, for example, a CNN network, in step 107. In step 108, the neural network is then learned, after which a weight vector or a weight file is created in step 109. The individual steps will be described in more detail below with reference to the example of Fig. 5 are shown: (1) Recording of vibration data, in the example ( Fig. 5) the amplitude of the acceleration of the supported load in the spatial direction x over time, determined on a vibration isolator (2) Generation of the spectrograms used here as input for the network. In the example, each spectrogram represents a process that is to be evaluated as "good" or "bad." (3) Labeling, ie assignment of the spectrograms to the defined classes “good” or “bad”. (4) Optionally, generation of variances by augmenting the spectrograms. This can be done manually, but preferably also computer-assisted. (5) The network used requires spectrograms for learning, testing, and validation during the learning process. The spectrograms are assigned to the previously defined classes; however, this assignment can also be done automatically programmatically. Sufficient spectrograms must be available for each class (e.g., 1,000 or 10,000 per class). (6) The network is created, in this example a CCN network. The network has two output neurons, one for the two defined classes ("good," "bad"). (7) The learning process is carried out. (8) The weights file is generated, which contains the parameterization of the individual elements and connections of the CNN network.

[0110] Fig. Figure 5 shows an example of the vibration behavior of an xy stage (“stage”) in the three spatial directions during different operating phases over a cycle. The upper graph shows the three amplitudes of the vibration behavior in the three spatial directions x, y, z over time during the different operating phases, starting with the acceleration in the spatial direction x, y, z measured by the acceleration sensor of a vibration isolator (only the acceleration in the x direction is used in the example) in the rest phase before the acceleration of the stage, followed by the acceleration measured by the sensor during the acceleration of the stage, then followed by the acceleration measured by the sensor during the uniform movement of the stage, then during deceleration, and in the rest phase after the movement of the stage. The lower graph shows the acceleration of the stage in the spatial directions x and y over time. It can be seen that in the example shown, an acceleration orbraking only occurs in the y direction, the vibrations or accelerations are measured in all spatial directions.

[0111] In the exemplary embodiment, it is clearly visible that each of the individual operating phases or cycles is characterized by a specific frequency response. The properties of this frequency response (frequencies, amplitudes, temporal sequence) depend on whether the machine is still delivering the required performance data or not. Furthermore, other information can also be obtained. For example, it is possible to detect errors in the system that affect the frequency response, such as a defective bearing, if this bearing affects the frequency response. This defective bearing can then lead to a "bad" classification; with an appropriate design of the neural network, it is even possible to directly identify the bearing in question as the source of the error.

[0112] For training, a sufficiently large number of spectrograms are generated, with each spectrogram corresponding to a complete run through the four operating phases, corresponding to a complete stage cycle. These are then "labeled," i.e., classified into the designated classes ("good" / "bad").

[0113] In Fig. 6 shows a spectrogram purely as an example. The spectrogram can contain the vibration data of the xy-table in one of the spatial directions x, y or z. However, it is also possible to superimpose the vibrations in all three spatial directions x, y and z in one spectrogram, as in the embodiment of Fig. 6 is the case, where Fig. 6, however, only shows the spectrogram in one spatial direction.

[0114] It is also possible to represent the oscillations in the spatial directions x, y and z in separate spectrograms and process them together in the artificial neural network.

[0115] Furthermore, the vibration behavior with respect to the rotation axes can be recorded and processed along the spatial directions x, y, or z. This allows more information to be used for classification, which can be useful and increases the reliability of the classification. However, additional information also increases the complexity of the neural network, so it is advisable to determine the information actually required for the "good" and "bad" classification in the specific application case.

[0116] The vibration data can also be represented in other ways, for example, using time-domain or frequency-domain representations. This is also assumed for the exemplary embodiments presented here and those that follow.

[0117] A spectrogram, as exemplified in Fig. 6, represents a collection of a large number of measured values. The spectrogram contains time courses for a large number of frequencies, thus also the spectral composition of the occurring oscillations. In a performance analysis, this large number of measured values, or their entirety, is evaluated and the result is expressed as an output value. This can be an indication of a class that, for example, characterizes the operating state as good or bad. Not only in the specific example of a spectrogram, but in general, the evaluation of the measured values ​​in their entirety takes place in the neural network, which typically also includes complex weighting and calculation of the measured values ​​in order to obtain the output value.

[0118] In the Fig. In the embodiment shown in Figure 7, an artificial neural network is to be trained to generate a classification into the classes "good" or "bad" from vibration data in the spatial direction x. For this purpose, three spectrograms of vibrations in the spatial direction x are shown purely as an example, each of which is classified into the classes "good" or "bad." It is understood that in practice, not just three spectrograms are used, but preferably a multiple of these, for example, 100 or more, or even 1,000 or more.

[0119] For classification, the weight file 13 generated during training is loaded into the network 15, and the spectrograms obtained during regular operation of the vibration isolation system 1, or more generally the measurement data 14, are fed into the network 15. Fig. 8. This data is then classified, i.e. assigned to at least two classes 16, 17, preferably into classes “good” or “bad”.

[0120] In a further, second, based on the Fig. 9 and Fig. In the embodiment illustrated in Figure 10, which is based on the aforementioned embodiment, the movement phases of acceleration, (uniform) movement, deceleration, and rest state of an xy-table are to be differentiated. This assignment enables a more precise classification of the operating state of the xy-table, ie, the classification can be made with greater reliability, since the vibration data differ in these movement phases, and not, as in the embodiment of the Fig. 5 assumed, over a complete movement cycle, including all four movement phases in a summed view.

[0121] The input data assumed are vibrations of a vibration sensor of a vibration isolation system 1 in a spatial direction, e.g. in spatial direction x, during the movement of an xy-table.

[0122] As a result, information about the quality of the setting or adjustment of the vibration isolation system 1 can be provided in the classes “good” and “bad”, separately for each movement phase.

[0123] This enables a more accurate performance analysis of the vibration behavior in the individual motion phases, so that deviations from the specification of the xy table or, more generally, of a mounted device or machine, for example a bearing failure as discussed above, can be detected even more accurately.

[0124] A sufficient number of spectrograms 20 are used for training, with each spectrogram corresponding to a specific movement phase (acceleration, movement, deceleration, rest position) or a specific combination of movement phases, such as the phase from the rest position to reaching a maximum speed. The spectrograms 20 represent the data of the movement quantity provided by the sensor 6a - 6d, which contain the temporal progression of the amplitudes of several oscillation frequencies. These spectrograms 20 are also divided into the designated classes, in the example into the classes "good" and "bad". Each spectrogram is additionally assigned information about the movement phase or the corresponding cycle of the xy table, as in the embodiment of the Fig. 9. Based on the four movement phases of the xy-table, the assignments in the Fig. 9 as follows: - Data P1: Data for movement phase 1, acceleration - Data P2: Data for movement phase 2, movement - Data P3: Data for movement phase 3, braking - Data P4: Data for movement phase 4, rest phase

[0125] For training, several spectrograms 20 of the different movement phases are used, which are separated according to the movement phases, Fig. 10.

[0126] For classification, the weight file 13 generated by training is loaded into the artificial neural network, and the measurement data 14, particularly in the form of spectrograms and the assignment of the movement phase from regular machine operation, are fed into the network 15. This data is then classified as "good" or "bad" and output according to the respective movement phases. Fig. 11.

[0127] According to yet another, third embodiment of the invention, additional information is to be used for performance analysis. Otherwise, this embodiment is based on the previously mentioned embodiments.

[0128] In the third embodiment, additional information is added or various pieces of information are combined. The input data can include, for example, the following information, although the following list is not intended to be exhaustive: - the vibration information of the vibration isolation system 1 and / or the supported load 4, in particular the supported machine or device - the vibration information of a connected component, such as a vacuum pump - the vibration information from the floor or the installation surface of the vibration isolation system 1, or the vibration information of a component within the machine, for example the vibrations of an optical sensor, or the vibrations of a cathode ray tube, or the vibrations of a vacuum pump - the strength of acoustic vibrations - the temperature of the environment, the vibration isolation system 1 and / or the supported load 4, in particular the supported machine or device - the strength of electromagnetic radiation on the machine or device.

[0129] The vibration information in turn relates to the vibration behavior, including the amplitude, the period, the frequency and / or the frequency spectrum.

[0130] Fig. 12 shows an example of a selection of input data, where the following assignments should apply: - X1: Vibration information of the vibration isolation system 1, - X2: Vibration information of the vacuum pump - Xn: measured values ​​of temperature, electromagnetic radiation, acoustic vibration

[0131] As a result, the generated output data, with appropriate combination of the input data, can provide information about the performance of the vibration isolation system 1 and / or the supported load 4, in particular of the supported machine or device, for example, operating parameters that lead to good or poor performance of the vibration isolation system 1 and / or the supported load 4, in particular of the supported machine or device. In the manner described, information about performance data can also be obtained - independent of vibration information. For example, electromagnetic radiation, which is characterized by a frequency of the radiation, and / or an amplitude of the radiation, and / or a direction of the radiation, can also be used to classify the performance data of the machine if the network is trained accordingly.

[0132] For training, a sufficient amount of information is required, which is “labeled” after appropriate processing, i.e. assigned to the given classes, Fig. 13.

[0133] The classification process based on measured values ​​or input data obtained during regular operation is shown purely by way of example Fig. 14.

[0134] According to yet another, fourth embodiment of the invention, the performance analysis of the vibration isolation system 1 is used to identify fault causes. Otherwise, this embodiment is based on the previously mentioned ones.

[0135] As a result, the generated output data, when combined with the input data accordingly, can provide information about whether the vibration isolation system 1 and / or the supported load 4, in particular the supported machine or the supported device, operates within the specified specification or whether deviations therefrom exist.

[0136] Information can be generated as to whether the vibration isolation system 1 or the machine or device is operating within specifications. If it is operating outside of specifications, the cause can be determined. This requires that the input data, or, during normal operation, the measured values, provide sufficient information to enable appropriate classification.

[0137] The input data may, for example, include information indicating faulty or suboptimally adjusted components or environmental conditions. The following list is exemplary and should not be considered exhaustive: - the vibration information of a vibration isolation system 1 operating within the specification, in particular of the machine or device - the vibration information of a vacuum pump operating within the specification - the vibration information of a vibration isolation system 1, in particular a machine or device, which is not operating within the specification - the vibration information of a vacuum pump that is not operating within the specification

[0138] The vibration information in turn relates to the vibration behavior, including the amplitude, the period, the frequency and / or the frequency spectrum.

[0139] Fig. 15 shows a selection of input data purely as an example, whereby the following assignments should apply - X1: Vibration information vibration isolation system 1 - X2: Vibration information vacuum pump

[0140] For training, several input data sets are used, representing the various combination possibilities. In the above example, this could be the following input data: - Vibration isolation system1 operates within specification, vacuum pump operates within specification - Vibration isolation system 1 does not operate within specification, vacuum pump operates within specification - Vibration isolation system 1 operates within specification, vacuum pump does not operate within specification

[0141] In addition to the data already discussed, the input data also includes information indicating a faulty component or environmental conditions. Further input data can therefore also relate to the supported load 4, in particular a supported machine or device, such as an xy table.

[0142] For example, a frequency spectrum that results from correct movement of an xy-table in a specific direction, such as the y-axis, can be used for training. If a bearing failure occurs in a bearing used for movement in the y-axis of this xy-table, a deviation in the frequency spectrum can occur during normal operation. Deviations in the frequency spectrum can also have a variety of causes, such as component aging, faulty / suboptimal repairs or software updates, changes in the machine's motion sequences for which it is not optimized, loose connections in cables, and compressed air fluctuations. Data from such components can also be used as input variables.

[0143] The training and generation of the “weights file” is carried out as described above, whereby a sufficiently large number of data records in the various categories must be ensured, Fig. 16.

[0144] The classification process based on data obtained during regular operation is shown purely by way of example Fig. 17. For classification, the "weights file" generated during the training of the artificial neural network is loaded into the network, and the data from the machine operation to be classified is fed into the network. This data is then classified and, if applicable, assigned to a faulty component.

[0145] In the above-mentioned embodiments, the sensors are arranged not only in or on the vibration isolation system 1, but also on or in the supported load 4, in particular on the plate or a machine supported thereon or on or in a supported device.

[0146] Furthermore, the sensors can also be arranged on or in the associated components, for example on an xy table, on an optical sensor, on light or electron optical elements of a microscope, on spectroscopic, scatterometric or interferometric measuring devices, or also on a scanning probe microscope, on a distance sensor, for example for measuring the distance of a light optical element from a substrate, on a casing, on a vacuum pump, or on another component of the vibration isolation system. The sensors may include, for example, optical sensors, ultrasonic sensors, acceleration sensors or distance sensors, with acceleration sensors being provided in the present case.

[0147] Additional sensors can be assigned to the environment of the vibration isolation system 1, for example, the floor or the installation surface, or also to the surroundings, for example, other components in the vicinity of the vibration isolation system. In particular, it is advisable to select components that can transmit vibrations to the vibration isolation system and / or to the supported load. This can also be, for example, the ceiling or a lifting device arranged in the surrounding area, such as an overhead or ceiling crane.

[0148] Magnetic field sensors or fluxgate magnetometers or saturation core magnetometers are used to detect electromagnetic fields.

[0149] Pressure sensors are provided to measure the strength of acoustic fields.

[0150] Thermocouples are provided to measure the temperature at the installation site and / or on components of the vibration isolation system and / or the supported load.

[0151] The generation of output data from the input data or the classification for performance analysis is carried out computer-assisted, whereby the programs can be stored in the computing device 3. Of course, it is also possible to store the programs or the components of the artificial neural network at other locations.

[0152] For example, in a further development of the invention, it is envisaged to store the components of the artificial neural network 10 centrally in a higher-level computing device and to use them for several decentrally arranged vibration isolation systems 1. This has advantages with regard to maintenance and training.

[0153] In a further development of the invention, the artificial neural network can also be designed in such a way that information or input data from several vibration isolation systems 1 are used to generate control parameters for the control and regulation of one or more vibration isolation systems 1 connected to the higher-level computing device after the performance analysis.

[0154] The generation of variances can be achieved by augmenting the spectrograms or other forms of representation of oscillations, i.e. by changing or transforming the data sets, i.e. intentional deviations of the spectrograms or input variables are generated - preferably also with computer support - in order to increase the number of different training data sets.

[0155] This is advantageous because a large amount of data is required for training, among other things to capture, as far as necessary, any variations in the input information that may occur during normal operation. Such data can also be generated artificially or with computer support within certain limits by artificially or computer-assistedly distorting the data. For example, noise can be added to the vibration data.

[0156] In this way, a multitude of spectrograms can be generated, which can be used to train the artificial neural network. The spectrograms are assigned to the desired categories or classes. According to one embodiment of the invention, the assignment can also be performed automatically with computer support, which can significantly simplify manual data processing.

[0157] For good performance analysis and diagnostics, it is helpful to have enough spectrograms available in the relevant classes. In certain cases, 10 to 100 input data sets or spectrograms per class may be sufficient, but it is better to have approximately 1,000 or even 10,000 or more data sets per class to improve and increase the quality of the performance analysis.

[0158] During regular operation of the vibration isolation system 1, a previously trained or learned artificial neural network is used. In an advantageous embodiment of the invention, this trained artificial neural network is further trained during ongoing operation.

[0159] In this way, the vibration isolation system 1 can be easily adapted to the conditions prevailing at the installation site.

[0160] In a further advantageous embodiment of the invention, it is provided that the artificial neural network uses learned data from a previous learning process or from another external source.

[0161] The performance analysis can generate information or output data, for example, control parameters for controlling or regulating the vibration isolation system 1 and / or the supported load 4. An important aspect of the invention is therefore that these control parameters can be used directly or indirectly to control or regulate the vibration isolation system 1 and / or the supported load 4.

[0162] The invention therefore also encompasses a method for controlling or regulating a vibration isolation system 1, wherein, during operation, control parameters for controlling the vibration isolation system 1 are obtained and used as part of a performance analysis as described above. In this way, a self-learning vibration isolation system 1 can be provided, which can, for example, react to changes in the environment, such as vibrations occurring via the floor.

[0163] The vibration isolation system 1 is designed as an active, stationary vibration isolation system and, in the exemplary embodiment, comprises four vibration isolators, whereby vibrations occurring during regular operation can be actively counteracted by means of the four actuators.

[0164] In a further aspect, the invention also includes a method for diagnosing, in particular for performance analysis, a vibration isolation system 1 comprising a vibration isolation system as described above.

[0165] The method can enable an analysis of the performance of the vibration isolation system, but also of the supported load.

[0166] In a further development of the invention, the artificial neural network is adapted. This may be necessary if influencing factors change during the network's use, e.g.: - more data can be generated when using the network, which can be used to improve classification - for example, a stored machine can better handle malfunctions due to improved software - the product being processed by the machine changes, which places different demands on the machine.

[0167] Since retraining the artificial neural network can be time-consuming, this advanced training proposes to continuously optimize the artificial neural network during regular operation. This saves training time and shortens the time between network updates. The process is presented in a schematic overview in Fig. 18. Here, the AI ​​model is continuously trained while applying it, using external performance indicators and evaluation indicators that classify the system's state. An AI model is understood here to be a model that contains the described neural network.

[0168] In addition to the architectures for classifying input data described above, other approaches exist. In the above embodiments, the input data is evaluated within a time period, for example, a time period of two seconds for a sequence consisting of the phases of acceleration, movement, deceleration, and rest.

[0169] There are also architectures that process a sequence of time segments. One example is LSTM (“Long Short-Term Memory”) neural networks, which can process a sequence of smaller time segments in a further dimension. Such architectures can be used in a similar way for the classification tasks described above. List of reference symbols 1 vibration isolation system 2a - 2d Vibration isolator 3 Computing device 4 Last 5a - 5d Actuator 6a - 6d Sensor 10, 11 set of spectrograms 13 Weight file 14 Measurement data 15 Network 16th, 17th grade 20 spectrograms 101 Step: Data acquisition 102 Step: Data processing 103 Step: Labeling 104 Step: Augmentation 105 training spectrograms 106 test spectrograms 107 Step: Creating a network 108 Step: Learning the network 109 Step: Creating a weight vector or weight file QUOTES CONTAINED IN THE DESCRIPTION

[0000] This list of documents submitted by the applicant was generated automatically and is included solely for the convenience of the reader. This list is not part of the German patent or utility model application. The DPMA assumes no liability for any errors or omissions. Cited patent literature

[0000] EP 2 295 829 B1

[0009] EP 3 726 094 A1

[0011]

Claims

[1] Vibration isolation system, in particular with diagnostic function, preferably stationary, active vibration isolation system, by means of which a vibration-isolated load can be supported, comprising at least one vibration isolator, preferably a plurality of vibration isolators, which can counteract vibrations occurring during operation by means of actuators, at least one sensor, preferably a plurality of sensors, which are configured to record measured values ​​relating to the vibration isolation system and / or the supported load and / or the environment and to provide them as input data to a computer-aided method for a performance analysis of the vibration isolation system and / or the supported load. [2] Vibration isolation system according to the preceding claim, wherein the computer-aided method comprises or is based on at least one artificial neural network. [3] Vibration isolation system (1), in particular according to one of the preceding claims, for the vibration-isolated mounting of a load, comprising at least one, preferably a plurality of vibration isolators (2a - 2d) and at least one sensor (6a - 6d), wherein the sensor (6a - 6d) is configured to detect a movement variable, in particular a mechanical vibration, movement or acceleration, and a computing device (3), wherein the computing device (3) contains an artificial neural network (10), wherein the artificial neural network (10) is configured to process data of the movement variable provided by the at least one sensor (6a - 6d) and to carry out a classification as to whether the vibration isolation system (1) with the mounted load is in an operating state with sufficiently little or too much disturbance by movements, and to output this classification. [4] Vibration isolation system according to one of the preceding claims, characterized by that the data of the movement quantity provided by the at least one sensor (6a - 6d) comprise spectrograms (20) which contain the temporal course of the amplitudes of several oscillation frequencies. [5] Vibration isolation system according to one of the preceding claims, wherein a trained or learned artificial neural network is used in regular operation. [6] Vibration isolation system according to one of the preceding claims, wherein the supported load is supported by three or four vibration isolators. [7] Vibration isolation system according to one of the preceding claims, wherein the supported load comprises processing machines or parts thereof, in particular in the semiconductor industry such as machines for processing semiconductor components and / or nanostructured substrates or elements, in particular wafers, masks or displays, lithography devices, or measuring devices or devices, in particular for measuring semiconductor components and / or nanostructured substrates, such as an electron beam microscope, or other sensitive devices, machines or systems, for example laboratory devices or medical devices, such as imaging examination devices such as magnetic resonance imaging (MRI). [8] Vibration isolation system according to one of the preceding claims, wherein the input data comprise the vibration behavior of the vibration isolation system, in particular the amplitude, the period, the frequency or the acceleration. [9] Vibration isolation system according to one of the preceding claims, wherein the input data comprise the temperature at the installation site, and / or at components of the vibration isolation system and / or the supported load. [10] Vibration isolation system according to one of the preceding claims, wherein the input data comprises the strength of electromagnetic fields at the installation site. [11] Vibration isolation system according to one of the preceding claims, wherein the input data comprises the strength of acoustic fields at the installation site. [12] Vibration isolation system according to one of the preceding claims, wherein the sensors are arranged in or on the vibration isolation system, and / or on the supported load and / or in the environment, in particular on the floor, the installation surface, or on components or elements in the environment of the vibration isolation system. [13] Vibration isolation system according to one of the preceding claims, wherein the artificial neural network forms at least one model which comprises at least one convolutional layer and / or at least one neural layer, and / or further components, for example activation layers. [14] Vibration isolation system according to one of the preceding claims, wherein the artificial neural network can be trained, in particular during operation of the vibration isolation system, and / or uses trained data from a previous training process. [15] Vibration isolation system according to one of the preceding claims, wherein the artificial neural network uses at least one of the methods of regression, machine learning or deep learning. [16] Vibration isolation system according to one of the preceding claims, wherein, for training the artificial neural network, data sets are modified or transformed with computer support in order to obtain a plurality of training data sets. [17] Vibration isolation system according to one of the preceding claims, wherein control parameters for controlling the vibration isolation system and / or the supported load are generated on the basis of the performance analysis. [18] Vibration isolation system according to one of the preceding claims, wherein the control parameters are used to control the vibration isolation system and / or the supported load. [19] Method for controlling or regulating a vibration isolation system, preferably according to one of the preceding claims, wherein during operation control parameters for controlling the vibration isolation system are obtained and used by the performance analysis. [20] Method for diagnosing, in particular for performance analysis, a vibration isolation system, preferably according to one of the preceding claims 1-14.

Images