Vibration isolation system with diagnosis function and diagnosis method

Abstract

The invention is based on the problem that the performance of a vibration isolation system is further improved in order to improve the performance of a machine or unit connected to the vibration isolation system, or that it can be evaluated in an improved manner. To this end, a vibration isolation system is provided, by means of which a load mounted in a vibration-isolated manner 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 an actuator; at least one sensor, preferably a plurality of sensors, configured to record measurements relating to the vibration isolation system and / or the installed load and / or the environment and to provide them as input data to a computer-aided method for performance analysis of the vibration isolation system and / or the installed load.

Description

Technical Field

[0001] This invention relates to a vibration isolation system and diagnostic method, preferably fixed. Specifically, the invention relates to an active, fixed vibration isolation system on which machines for processing semiconductor devices and / or nanostructure substrates or components, particularly wafers, masks, or displays such as flat panel displays, are arranged, or machines for measuring semiconductor devices and / or nanostructure substrates, such as electron microscopes.

[0002] In addition, laboratory equipment and / or medical equipment, such as imaging equipment, such as magnetic resonance imaging (MRT) scanners, or microscopes, can also be placed on this fixed vibration isolation system. Background Technology

[0003] Active, fixed vibration isolation systems are used, particularly in the semiconductor industry, to support processing machinery (such as lithography equipment) and measuring equipment (such as electron microscopes) in order to provide vibration-isolated support for sensitive equipment used to process semiconductor components.

[0004] This vibration isolation system for machines or machine parts typically consists of a plate supported in a vibration-damping manner on at least three, usually four, and sometimes more than four isolators, each of which contains a spring.

[0005] Alternatively or additionally, one or more vibration isolators may be used, not arranged on a plate but between two components in the equipment, such as between the base frame and the platform or granite slab. Vibration isolators may also be arranged outside the machine; for example, the entire machine may be placed on one or more vibration isolators.

[0006] This protects the machine or its components from vibration. The machine or its components, together with any plates present, form a vibration-damping load.

[0007] Active vibration isolation systems are particularly known in practice, where the isolator includes actuators, especially magnetic actuators, in addition to springs, which actively cancel out vibrations. Vibration isolation systems also exist where additional sensors or actuators are attached as independent units to other parts of the machine, separate from the isolator. Vibrations are detected by sensors, which can be located either at and / or near the base of one or more isolators, or on the vibration-isolated load. Based on the sensor signals, the actuators are controlled to compensate for the vibrations by generating reaction forces.

[0008] This active control can be used to counteract vibrations entering the system from the outside, as well as vibrations generated by the vibration isolation load itself (e.g., vibrations generated by movable platforms (such as platforms) or mobile robots).

[0009] The applicant's patent specification EP 2 295 829 B1 illustrates a vibration isolation system in which an actuator is provided to generate compensating forces in multiple degrees of freedom.

[0010] This vibration isolation system includes a control unit (often called a controller) for processing sensor signals. The control unit detects the sensor signals and generates a control signal based on the sensor signals, which controls the actuator to generate a compensating force.

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

[0012] The present invention is based on the task of further improving or better evaluating the performance of a vibration isolation system, thereby improving or better evaluating the performance of a machine or equipment connected to the vibration isolation system, or optimizing the availability of the machine.

[0013] In particular, the present invention is based on the task of obtaining further data on vibration isolation systems and / or vibration-isolated machines or equipment, which, for example, can determine or predict the performance of the machine or equipment, or enable early detection of changes on the machine or in its environment, such as signs of wear, and provide information for remedial action. In this regard, machine performance should be understood as its precision, accuracy, availability, and productivity.

[0014] The inventor undertook this task.

[0015] This task is solved by a vibration isolation system with diagnostic capabilities as described in one of the independent claims, and a method for diagnosing the vibration isolation system and / or loads supported on it in a vibration-isolated manner. Preferred embodiments and further embodiments of the invention can be found in the corresponding dependent claims.

[0016] Therefore, the subject matter of the first aspect of the present invention is:

[0017] A vibration isolation system, particularly a fixed active vibration isolation system with diagnostic capabilities, is used to support a vibration isolation load, comprising: At least one vibration isolator, preferably multiple vibration isolators, is provided, which can counteract vibrations occurring during operation by means of an actuator. At least one sensor, preferably multiple sensors, is designed to record measurements related to the vibration isolation system and / or supporting load and / or environment, and to provide these measurements as input data to a computer-aided method for performance analysis of the vibration isolation system and / or supporting load.

[0018] Vibration isolators can actively or passively counteract vibrations. Active and passive vibration isolators can also be used together to form components of a vibration isolation system.

[0019] Measurements related to the supporting load may specifically relate to the machine or its components supported by a vibration isolation system. For example, a vibration sensor may be attached to an optical sensor of such a machine, or, in the case of an electron microscope, to an electron beam column. The sensor does not even need to be attached to the supporting load or the vibration isolation system itself. For example, vibrations of the system, the load, or its components may also be detected by an optical sensor located outside the system.

[0020] For the purposes of this disclosure, performance analysis should be understood as a classification of operating conditions regarding harmful disturbances and effects. In vibration isolation systems, these disturbances are particularly those caused by undesirable system motion, especially vibrational motion. In addition to the classification of disturbance variables, performance analysis also includes the classification of whether the system operates well, particularly optimally, including whether the system has been optimally tuned or set up. For example, performance can be classified as good or poor based on system components, such as the settling time after the xy stage moves, or the frequency response that occurs during this process, or other environmental conditions, such as other processes or operating conditions occurring simultaneously in the machine, temperature, or the load on the stage.

[0021] Generally, for performance analysis, a distinction can be made between operating states and process or machine functions. In both cases, forces from the machine's actuators or the environment are introduced into the dynamic system, the machine's structural dynamics. In structural dynamics, vibrations are excited and can be detected by sensors in a vibration isolation system. In operating states, quasi-static force introduction or constant vibration is assumed. In process or machine functions, transient force introduction or vibration states exist. In both cases, sensors in the vibration isolation system can detect and evaluate the amplitude and spectral composition of the vibration signal. In the case of process or machine functions, the temporal behavior and spectral composition of the vibration amplitude are also considered. From these observations, the corresponding operating states and process or machine functions can then be evaluated and classified as part of the performance analysis.

[0022] Specifically, performance analysis includes an evaluation of the numerous measurements provided by the vibration isolation system's sensors, as well as an assessment of the system's operating status, expressed as output values. Additional or other information, such as whether the machine is producing the desired quality, may also be used or utilized. Output values ​​may contain one of several categories characterizing the system's operating status. In the simplest case, there are only two categories, characterizing good and poor operating conditions. Output values ​​can also be sliding values ​​within a specified range, for example, as a percentage. A 100% output value could represent optimal operating conditions, while a 75% value could, for example, represent a barely acceptable state.

[0023] Motion, such as vibration and its suppression, is naturally of particular importance to the performance of vibration isolation systems. Therefore, according to a second aspect of this disclosure, a vibration isolation system for supporting a load in a vibration-isolated manner is provided, comprising at least one, preferably multiple, vibration isolators and at least one sensor and a computing device, wherein the sensor is arranged to detect motion variables, particularly mechanical vibration, motion, or acceleration, and wherein the computing device includes an artificial neural network designed to process motion variable data provided by the at least one sensor and classify whether the vibration isolation system and the supported load are in an operating state with sufficiently small or excessive motion disturbance. This classification can then be output. In a simple case, this output is intended to notify the operator or monitoring personnel of the operating state. This output can also be used by the vibration isolation system itself or by another device connected via information technology to influence the operating state when it is classified as being subject to excessive motion disturbance. The motion variables as input data are primarily vibrations, particularly their amplitude and frequency. However, acceleration and velocity may also be used alternatively or alternatively. For simplicity, the aforementioned operating state is also referred to below as “good” (operating state with sufficiently small motion disturbance) or “bad” (operating state affected by excessive motion disturbance). Of course, as mentioned above, classification can include further or more refined hierarchies, such as in the form of sliding output values.

[0024] A particularly preferred form of data output from a sensor and processed by an artificial neural network is a spectrogram. For the purposes of this disclosure, a spectrogram should be understood as a dataset that specifically contains the time progression of amplitudes at multiple vibrational frequencies.

[0025] This invention enables the use of sensor measurements as input data to obtain performance data of vibration isolation systems and / or load-bearing systems.

[0026] Support loads may include, for example, processing machines or components thereof, particularly in the semiconductor industry, such as machines for processing semiconductor devices and / or nanostructure substrates or elements, especially wafers, masks, or displays; lithography equipment; or measuring equipment or devices, particularly for measuring semiconductor elements and / or nanostructure substrates, such as electron microscopes; or other sensitive equipment, machines, or systems, such as laboratory equipment or medical devices, such as imaging examination equipment like magnetic resonance imaging (MRI) scanners. For the sake of brevity, the terms “machine” or “equipment” will generally be used below.

[0027] In many embodiments of the invention, the supporting load may comprise a plate, particularly as a mounting plate, on which a machine or equipment may be arranged and fixed. The equipment or machine arranged on such a plate is thus protected from vibration. The plate and the equipment or machine arranged thereon can together constitute a vibration-isolated load. The supporting load may be supported by three points via three, but may also be supported by four or more vibration isolators.

[0028] According to a preferred embodiment of the invention, the sensor is designed to record various information or measurements about the vibration isolation system and / or supporting load, and even the environment, during operation, and provide them to a suitable computer-aided process. According to the invention, this input data can be used for performance analysis of the vibration isolation system and / or supporting load. The results of the performance analysis can be used to diagnose the vibration isolation system and / or supporting load, thereby enabling the vibration isolation system to have diagnostic capabilities.

[0029] According to a preferred embodiment of the invention, the sensor can provide a wide range of information, such as measurements that can be viewed in a specific context, which are only of limited use or even impossible to evaluate using rule-based or statistical methods. This can be of great significance when used with vibration isolation systems on or with machines or systems in the semiconductor industry or in the field of inspection or measurement technology, or when used generally on or with machines.

[0030] For example, for certain types of machines, minimizing or eliminating vibration as much as possible may be advantageous. Performance, such as the optical resolution of a measurement or inspection system, may depend heavily on the vibration that occurs.

[0031] Therefore, there may be a complex relationship between the vibrations that occur and the performance of the load supported by the vibration isolation system.

[0032] To reduce vibration, sensors capable of measuring vibrations occurring at specific points in the system can be installed. These sensors can be, for example, accelerometers capable of detecting acceleration in one of the x, y, or z directions in a Cartesian coordinate system. Since, in most cases, the performance of a vibration isolation system, particularly concerning the performance of the supported machine or equipment, is related to the supporting load, and since the simultaneous vibrations can also be measured, statements about the performance of the supporting load or the supported machine or equipment can be made based on sensor measurements.

[0033] The inventors discovered that further information can be obtained from sensor readings. For example, it can be determined whether the supporting machine is still operating within specifications. Vibration characteristics can also be used to determine possible causes of deviations from specifications, such as whether the installed XY stage or cross stage (“stage”) is functioning properly, or whether, for example, a vacuum pump connected to the system is transmitting excessive vibration to the system, or whether there is an earthquake-related cause for the deviation. For the purposes of this disclosure, vibration characteristics should be understood as the temporal variation of a spectral distribution and the correlation between signals, where one signal represents the excitation and another signal or signals represent the response behavior. In general, vibration characteristics can also be represented by the time-correlated intensity of various frequencies of the spectrum that appear, for example, after a pulse excitation.

[0034] Furthermore, information about the machine's operating status can be obtained from vibration characteristics, such as whether the XY stage is in a specific position, like the position used for substrate transfer, or whether it is still in motion. The vibration spectrum may change depending on the stage's position. Specific vibration patterns can also indicate specific operating conditions or malfunctions, such as vacuum pump defects.

[0035] In addition, information about machine wear or maintenance status can be obtained from vibration characteristics. This information can also be provided to machine control software or operators.

[0036] Current operating conditions can also be considered, such as the idling state of the supporting load, acceleration, or the motion of the xy stage.

[0037] Other measurable factors that affect machine performance include environmental vibrations, such as floor vibrations, especially at the installation location, or the presence of electromagnetic fields, or temperature.

[0038] Therefore, the subject of this invention is to utilize information obtained from sensors in a vibration isolation system not only for direct vibration isolation, i.e., controlling actuators to reduce vibration, but also for further performance analysis of the vibration isolation system and / or the loads supporting it, and to perform performance analysis capable of diagnostics in this way.

[0039] According to one embodiment of the invention, this information is also intended for use, for example, in controlling and regulating the vibration isolation system and / or the load or machine or equipment supported on it, or other system components such as a vacuum pump.

[0040] The evaluation of measurements can be very complex. For vibration-related data, such as different frequencies with varying amplitudes, it can be represented in the time domain, frequency domain, or spectrum. Other environmental conditions or machine conditions that need to be considered further increase the complexity.

[0041] Rule-based computer-aided programs and methods can be used for evaluation or performance analysis. Certain rules can be defined to generate preliminary information based on measurements or data. An example of a simple rule is that if the amplitude of vibration exceeds a specified limit, the machine is no longer operating within the specified range.

[0042] Therefore, analytical, statistical, or other mathematical methods can be used to generate the required information based on the analytical or statistical rules to be defined. These rules are usually generated manually.

[0043] This makes it possible to obtain information for simple performance analysis, which may include, for example, the following: - Earthquake or floor vibration, - "Window Functions": Whether the xy stage is in the transfer position. - Monitor vibration levels according to limits.

[0044] When input data becomes too complex or interdependent, statistical or other rule-based methods that require manually creating and storing rules to generate the desired information from the input data reach their limits.

[0045] For example, situations may arise where vibration or acceleration exceeds limits if it occurs under specific operating conditions of the machine serving as a supporting load. In such cases, the short-term excess may be irrelevant or almost irrelevant to performance, and / or if there is a compensating effect on the machine, and / or if the excess does not occur frequently. Manually defining rules for such complex relationships is not only extremely time-consuming but also highly error-prone, and in cases of complex relationships, it may be impossible or at least unreliable.

[0046] Therefore, according to the present invention, a computer-aided performance analysis based on an artificial neural network is provided, which can be trained rather than using fixed rules. This allows for significantly more information about the vibration isolation system and / or the performance of the supporting load, which can then be used for control and regulation.

[0047] In one aspect of the invention, a vibration isolation system is provided that uses artificial intelligence, preferably an artificial neural network, method or process to perform performance analysis of the vibration isolation system and / or supporting load.

[0048] Input data obtained from measurements may relate to the vibration isolation system itself, support loads, or other information. This input data can be used individually or in combination. For diagnostic purposes, information can be categorized individually or in combination.

[0049] Suitable sensors can be selected and provided to record the required measurements. If a specific sensor is mentioned below, this is merely an option and should not be considered exhaustive.

[0050] The input data may, for example, relate to the vibration behavior of the vibration isolation system, including signal distribution in the spectrum (variations in the frequency range) and impulse and decay behavior (variations in the time range). Suitable sensors may include, for example, accelerometers, preferably for measuring three spatial directions.

[0051] To record basic measurements, sensors can be mounted on or within the vibration isolation system and measure vibration behavior, for example, in one or more directions. Single-degree-of-freedom sensors are preferred. By using multiple such sensors, multiple degrees of freedom, such as up to six, or vibration directions, can be recorded accordingly. According to another embodiment, one or more sensors capable of measuring more than one direction or more than one degree of freedom may also be used.

[0052] Therefore, the sensor can be arranged not only in or on the vibration isolation system, but also on the supporting load, such as in or on the supported machine, or in or on the supported equipment. If the supporting load includes a plate, it may also be advantageous to assign the sensor to the plate according to a preferred embodiment of the invention.

[0053] In addition, sensors may also be placed on or within associated components, such as on or within the xy stage, optical elements, cover plates, vacuum pumps, or other components of vibration isolation systems.

[0054] Input data may also include the operating status ("on", "off", "idling", "running", "under maintenance", "faulty") of the vibration isolation system and / or supporting load. This provides information about the static or operational status of the vibration isolation system and / or supporting load, possible accelerations, the motion of the xy stage, or the operating status of components (such as a vacuum pump). For this purpose, optical sensors, ultrasonic sensors, or distance sensors can be used, for example.

[0055] Additional sensors can be assigned to the environment of the vibration isolation system, such as the floor or installation site, or to the surrounding environment, such as other components near the vibration isolation system. In particular, it is recommended to select components that can transmit vibrations to the vibration isolation system and / or support loads.

[0056] Furthermore, recording the intensity and spectral composition of the dominant electromagnetic field at the installation site of the vibration isolation system can also be helpful, as these fields can impair the performance of the vibration isolation system and / or the supporting load. In the context of this invention, information other than vibration can also be used to determine machine performance data. For example, the magnetic field spectrum can be used instead of vibration data to determine whether the machine is still operating within specifications, i.e., within permissible tolerances. Magnetic field and vibration data can also be combined for this purpose. Although this becomes increasingly complex, this is precisely where the advantage of the device described herein lies—its ability to handle complex relationships. In most cases, the performance of the supporting load depends on various environmental factors, such as magnetic fields. By recording various signals, the effects of different combinations of interference on machine performance can also be considered. Electromagnetic fields can be measured using magnetic field sensors, fluxgate magnetometers, or saturated core magnetometers.

[0057] Measuring the intensity of the sound field present at the installation location can also be advantageous. Similar advantages apply here to the embodiments described above for detecting magnetic fields. In most cases, the performance of the supporting load depends on various environmental factors, such as sound pressure. By detecting various signals, the effects of different combinations of interference on machine performance can also be considered. For this purpose, pressure sensors and microphones can be installed, for example.

[0058] Furthermore, for the same reason, obtaining temperature information at the installation location and / or at vibration isolation system components and / or supporting loads may also be advantageous. Appropriate thermocouples can be installed for this purpose.

[0059] In most cases, the performance of the support load depends on various environmental factors, such as temperature. By recording various signals, the impact of various combinations of disturbances on machine performance can also be considered. Thus, for example, it may be possible to detect faulty operating conditions of the vibration isolation system or support load, such as overheating of a faulty component, or to detect operating conditions where normal operation is no longer possible, for example, in the case of optical measurements.

[0060] Based on the aforementioned measurement variables, it is clear that any correlation between any input variable and machine performance data can be determined, as long as it is available and measurable, without being limited to specific measurement variables and their sensors. Therefore, a further aspect of the invention specifies that the vibration isolation system is configured to perform a classification of the performance status of the equipment supported by the vibration isolation system based on the correlation of input data from multiple sensors recording different measurements.

[0061] According to the present invention, performance analysis can be performed based on input data using computer-aided methods, particularly with the support of artificial neural networks, to classify the performance of vibration isolation systems and / or supporting loads.

[0062] The classification may include, for example, characteristics of "good" or "poor". Based on this performance analysis, the diagnosis may provide, for example, an assessment of "operable" or "inoperable". The "good" or "poor" classification may also relate to the production capacity of equipment supported by a vibration isolation system. If the settling time after equipment movement is too long, and the process performed by the equipment depends on whether the equipment is stationary, an unfavorable vibration spectrum can lead to downtime, thereby slowing down production or measurement. On the other hand, if the process does not directly take vibration into account, vibration can affect measurement accuracy or production accuracy. In this regard, the condition of the vibration isolation system can also be classified as, for example, "good", "acceptable", or "poor". Further diagnosis could be "operable with limitations" or "operable after changes to the control or adjustment of components".

[0063] Therefore, diagnosis may involve, for example, the setup or adjustment of the vibration isolation system itself, where the vibration isolation system's specifications can be used as reference values. A "good" classification may indicate that the vibration isolation system is operating within the specified range ("operable").

[0064] In a further embodiment of the invention, the support load can be analyzed to determine, for example, whether the supported machine or equipment is operating or capable of operating within specifications (“good”) or is not operating (“bad”). Here, the specifications of the machine or equipment can also be used as classification criteria. All of the above signals can be used as input information for classification, such as environmental specifications (e.g., floor vibration levels), or electromagnetic or acoustic environmental conditions and other vibration data, wherein various data can also be combined for classification.

[0065] According to the present invention, computer-aided methods, preferably including or based on artificial neural networks, can be used to generate output data from input data. Suitable data preparation or processing methods (not described in detail here) can be used to prepare measurements or generate input data.

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

[0067] In a preferred embodiment of the invention, the artificial neural network can be trained to include at least one model comprising at least one convolutional layer and / or at least one neuron layer. It also preferably includes at least one additional component: an activation layer, a pooling layer, a flattening layer, and / or a dropout layer. Advantageously, the artificial neural network contains data that weights at least one, preferably multiple, and particularly preferably all weightable components of the model. This weighting can be determined particularly easily through the training process.

[0068] Here is a brief overview of the possible layers of a suitable artificial neural network: - Input layer: Here, the spectrogram is read into the network. - Convolutional level or convolutional layer: Many configurations are possible (e.g., regarding the number of layers, the number and size of filters, activation functions, max pooling functions).

[0069] - Flattening layer: Converts the 3D result of the last convolutional layer into a 1D format for transmission to the neural layers. - Neural layers: Various configurations are possible (e.g., regarding the number of layers, number of neurons, activation function, dropout). In some cases, neural layers can be omitted entirely (this also applies to classification layers). In this case, classification can also be generated by convolutional layers.

[0070] - Neural layers used for classification: Neurons represent the desired categories; for example, two neurons represent the categories "good" and "bad". During learning, network parameters are optimized by iteratively comparing the generated target / actual values ​​for classification; these parameters are then applied in the classification application. Softmax activation can be used to generate floating-point values ​​to represent the membership of the spectrogram to the "good" and "bad" categories.

[0071] The learning process of this artificial neural network is often understood as an adaptive algorithm or "machine learning". If a series of layers or hierarchies are used, it is also called "deep learning", although there is no precise definition for the number of layers required.

[0072] The following is a brief overview of the training or learning of artificial neural networks, using input data related to the vibration behavior of a vibration isolation system as an example.

[0073] Vibration frequency is used as a measurement variable, recorded by appropriate sensors, and provided as an input variable to the artificial neural network in the form of spectrograms. Each spectrogram can represent a process to be evaluated as "good" or "poor," provided that these two categories are formed for the analysis. Then, "labeling" is performed, assigning the spectrograms to the "good" or "poor" category.

[0074] Variations can be generated by enhancing the spectrogram, i.e., by modifying or transforming the dataset. This means that it is preferable to also use computers to introduce intentional biases in the spectrogram or input variables to increase the number of different training datasets.

[0075] In this way, by recording a large number of localization sequences and possible augmentations, a large number of spectrograms can be generated for training the artificial neural network used. For this purpose, the spectrograms are assigned to predetermined categories or classifications. According to one embodiment of the invention, this assignment can also be performed automatically by a computer.

[0076] For effective performance analysis and diagnostics, it is helpful to have a sufficient number of spectrograms available for the relevant category. In some cases, 10 to 100 input datasets per category, such as spectrograms, may be sufficient, but it is best to provide, for example, about 1,000 or even 10,000 or more datasets to improve the quality of performance analysis.

[0077] Artificial neural networks are created, for example, as convolutional networks as described above. In the example discussed, the artificial neural network may have two output neurons, representing two defined categories, "good" and "bad." This allows for learning. The training of artificial neural networks is generally known to experts and will not be elaborated upon here. The result is the creation of a so-called "weight file," which contains parameterizations of the individual elements and connections of the CNN network.

[0078] It goes without saying that the training method described above is merely an example; the categories, as well as the number and type of input variables, may vary, and other architectures of artificial neural networks and other input data can also be used. For example, 4-dimensional input information can also be used, where the fourth dimension can be a time axis, with the times of different spectrograms located on it.

[0079] It should also be understood that, in the routine operation of vibration isolation systems, it is preferable to use pre-trained or learned artificial neural networks.

[0080] In an advantageous embodiment of the invention, the artificial neural network is specified to be trained or further trained during the normal operation of the vibration isolation system. This provides the advantage that the vibration isolation system according to the invention can be easily adapted to the conditions commonly found at installation sites.

[0081] In a further advantageous embodiment of the invention, the artificial neural network is specified to use training data from a previous learning process or from another external source.

[0082] Performance analysis can be used to generate information or output data, such as control parameters for controlling or regulating vibration isolation systems and / or supporting loads. Therefore, an important aspect of this invention is that these control parameters can be used directly or indirectly to control or regulate vibration isolation systems and / or supporting loads.

[0083] Therefore, the present invention also includes a method for controlling or regulating a vibration isolation system, wherein control parameters are obtained during operation as part of the performance analysis described above and used to control the vibration isolation system. This provides a self-learning vibration isolation system that can, for example, respond to environmental changes, such as vibrations occurring through the floor.

[0084] Vibration isolation systems may include active, fixed vibration isolation systems as described above. In this case, the load of the vibration isolation support may be supported by multiple isolators, and the vibration may be actively counteracted by multiple actuators.

[0085] Another aspect of the present invention includes a method for diagnosing a vibration isolation system comprising the vibration isolation system described above.

[0086] Another aspect of the invention also includes a processing machine, particularly in the semiconductor industry, such as a machine for processing semiconductor devices and / or nanostructure substrates or elements, particularly for processing wafers, masks or displays, lithography equipment, or measuring equipment or apparatus, particularly for measuring semiconductor elements and / or nanostructure substrates, such as an electron microscope, or another sensitive device, machine or system, such as laboratory equipment or medical equipment, such as imaging examination equipment, such as a magnetic resonance computed tomography (MRT) scanner, including the vibration isolation system described above.

[0087] This method can be used to perform performance analysis on vibration isolation systems and also on supporting loads.

[0088] Further details of the invention will be apparent from the description of the illustrated embodiments and the appended claims. Attached Figure Description

[0089] Figure 1 A schematic top view showing the basic structure of a fixed vibration isolation system. Figure 2 This diagram illustrates the structure and usage examples of using machine learning with the help of artificial neural networks to evaluate input data. Figure 3 A schematic diagram showing possible structures of suitable artificial neural networks is provided. Figure 4 A schematic diagram illustrating the process of training an artificial neural network is shown. Figure 5 The vibration behavior of the xy stage in the three directions of the Cartesian coordinate system during different operating phases is shown. Figure 6 An example of a spectrum diagram is shown. Figure 7 This illustrates an example of the process of training an artificial neural network using spectrograms according to the first embodiment. Figure 8 Showing according to Figure 7 The illustrated embodiment is an example of the process of classification using the obtained spectrograms during normal operation. Figure 9 An example of a spectrum diagram according to the second embodiment is shown. Figure 10 An example of the process of training an artificial neural network using a spectrogram according to the second embodiment is shown.

[0090] Figure 11 Showing according to Figure 10 The illustrated embodiment is an example of the process of classification using the obtained spectrograms during normal operation. Figure 12 An example of input data according to the third embodiment is shown. Figure 13 This illustrates an example of the process of training an artificial neural network using input data according to the third embodiment. Figure 14 Showing according to Figure 13 Example of the classification process in the embodiment, Figure 15 An example of input data according to the fourth embodiment is shown. Figure 16 This illustrates an example of the process of training an artificial neural network using input data according to the third embodiment. Figure 17 Showing according to Figure 15 Examples of classification processes in the embodiments, and Figure 18 A schematic diagram is shown of the process used to adjust an artificial neural network. Detailed Implementation

[0091] In the following detailed description of preferred embodiments, for clarity, the same or substantially the same components are designated by the same reference numerals in these embodiments. However, for better illustration of the invention, the preferred embodiments shown in the drawings are not always drawn to scale.

[0092] Figure 1 A schematic top view showing the basic structure of the vibration isolation system 1 is shown. The vibration isolation system 1, also schematically shown, includes multiple isolators 2a-2d, each containing springs and supporting the vibration isolation load 4. Figure 1 As shown in the example, the load 4 can be supported by three-point support via three or four vibration isolators 2a-2d.

[0093] In the example shown, for clarity, support load 4 consists only of the plate; the machine or equipment supported in a vibration-isolated manner is not shown. In other words, in normal operation, support load typically includes, in addition to the plate, the machine or equipment to be supported in a vibration-isolated manner.

[0094] The support load 4 may include, for example, processing machinery or components thereof, particularly in the semiconductor industry, such as machines for processing semiconductor elements and / or nanostructure substrates or elements, particularly machines for processing wafers, masks or displays, lithography equipment, or measuring equipment or devices, particularly for measuring semiconductor elements and / or nanostructure substrates, such as electron microscopes, or other sensitive equipment, machines or systems, such as laboratory equipment or medical equipment, such as imaging inspection equipment like magnetic resonance imaging (MRI) scanners.

[0095] Each of the vibration isolators 2a-2d includes actuators 5a-5d and sensors 6a-6d. Sensors 6a-6d detect vibrations of the vibration isolation support load 4 and / or vibrations of the isolator base coupled to the ground. Like the actuators 5a-5d, sensors 6a-6d are connected to a computing device 3 centrally arranged in this example. Based on the signals from sensors 6a-6d, computing device 3 calculates compensation signals for controlling actuators 5a-5d. In this case, computing device 3 also acts as a control unit.

[0096] Vibration isolators 2a-2d are preferably effective in both the vertical and horizontal directions, as in this embodiment. Vibration isolators 2a-2d may include springs, particularly pneumatic springs. The sensor, spring, and actuator can be integrated components of such a vibration isolator or designed as independent units. In the latter case, the independent, particularly spatially adjacent, units of the sensor, spring, and actuator together constitute a logical unit with vibration isolator functionality, acting not only as a passive spring but also actively canceling vibrations.

[0097] The vibration isolation system 1 according to the invention includes a plurality of actuators that actively counteract vibration by generating forces acting on a load in a vibration-isolated manner. The actuators may be designed, in particular, as in this embodiment, as magnetic actuators, or as pneumatic actuators. Control parameters for controlling or adjusting the vibration isolation system 1 may be converted into current, for example, in the form of control signals, via a digital-to-analog converter (not shown), to control the magnetic actuators.

[0098] The computing device 3 is designed to execute a program that generates control signals to produce compensating forces, taking into account sensor measurements. The control signals are based on sensor signals used to control actuators and generate reaction forces. Preferably, the compensating forces are generated in at least two, particularly preferably at least three, degrees of freedom via actuators in an isolator.

[0099] According to a preferred embodiment, Figure 1 The vibration isolation system 1 shown includes actuators 5a-5d, through which compensating forces can be generated in three translational degrees of freedom according to three spatial directions x, y and z, and / or in three rotational degrees of freedom.

[0100] The vibration isolation system 1 according to the invention can be designed as a fixed, active vibration isolation system, meaning that it stands firmly on a designated mounting surface during operation and actively counteracts vibrations that may be transmitted, for example, through the floor. In another embodiment, the vibration isolation system is constructed as passive. In this case, the computing device is designed specifically to monitor the operating status of the system with supporting loads using performance analysis based on sensor input data.

[0101] The sensor measurements are provided as input data to a computer-aided method stored in computing device 3, and are used not only for vibration reduction, but also for performance analysis of vibration isolation system 1 and / or supporting load 4. This allows performance data of vibration isolation system 1 or supporting load 4 to be obtained.

[0102] Sensors 5a-5d are configured to record various information or measurements about the vibration isolation system 1 and / or the supporting load 4, and even the environment (e.g., the mounting surface), in order to perform performance analysis of the vibration isolation system 1 and / or the supporting load 4 as part of diagnostics. For simplicity, only sensors 5a-5d associated with the plate are shown in the illustrated embodiment. Additional sensors may be associated with the supported equipment or machine and / or the environment, such as the floor or mounting surface, walls, ceiling, etc.

[0103] According to the present invention, a computer-aided performance analysis based on an artificial neural network is provided, which can be trained rather than based on fixed rules. Figure 1 For illustrative purposes, the artificial neural network 10 is schematically shown as a component of the computing device 3 (specifically designed as a control unit).

[0104] Therefore, the present invention provides a vibration isolation system 1 that uses artificial intelligence methods, particularly artificial neural networks, for performance analysis and diagnosis of the vibration isolation system 1 and / or the supporting load 4.

[0105] The input data obtained from sensor measurements may relate to the vibration isolation system 1 itself, the supporting load 4, or other information. This input data can be used individually or in combination. For diagnostic purposes, the information can be categorized individually or in combination.

[0106] exist Figure 1 In the illustrated embodiment, the input data includes the vibration behavior of the plate. This may include the plate's amplitude, period duration, frequency, and / or frequency spectrum. For this purpose, sensors 5a-5d are designed as accelerometers. In this example, the acceleration data in the x-direction from accelerometer 6a of isolator 2a is used here.

[0107] The following is an example of an embodiment of a vibration isolation system 1 according to the present invention, which is designed to classify the quality of the setup or adjustment of the vibration isolation system 1 as "good" or "poor". It should be understood that these two categories are given only as examples, and other categories such as "moderate" or other classifications are possible and conceivable. For this embodiment, it is assumed that the supporting load 4 is a vibration isolation plate having an xy stage ("stage"). This xy stage allows movement in the x and y spatial directions.

[0108] It will be obvious to those skilled in the art that the xy stage specified in the example is an example chosen for illustrative purposes only, and its basic principles and methods are applicable to other installed machines and equipment.

[0109] Figure 2 The example illustrates the structure and use of machine learning to evaluate input data using an artificial neural network. From input data, such as the amplitude of an oscillation over time or the acceleration representing the oscillation, a spectrogram can be generated through data processing—in this example, a time-series frequency spectrum—which can then be fed into the artificial neural network. The output data is a classification, assigning the spectrogram to either a "good" or "bad" category.

[0110] Figure 3 A schematic representation of a suitable artificial neural network structure is shown. Numbers represent different levels or layers, which are only briefly mentioned below for simplicity.

[0111] (1) Input layer (2) Convolutional level or convolutional layer (3) Flattening layer (4) Nerve layer (5) Neural layers used for classification.

[0112] In a suitable embodiment of the invention, the artificial neural network forms at least one model comprising at least one convolutional level (convolutional layer) and / or at least one neuron level (neural layer).

[0113] The artificial neural network is fed with data (in this example, a spectrogram) to weight at least one, preferably multiple, and particularly preferably all weightable components of the model. This weighting can be determined particularly easily through the training process.

[0114] Figure 4 This diagram schematically illustrates a suitable and possible process for training an artificial neural network, where the output data contains two categories. The process includes steps indicated by reference numerals 101 to 109 in the flowchart. In step 101, data is recorded. Step 102 is data processing. In step 103, the processed data is labeled. Then, in step 104, augmentation is performed to generate a training spectrogram set (step 105) and a test spectrogram set (step 106). This is used in step 107 to create a network, such as a CNN network. In step 108, the neural network is trained, followed by the creation of weight vectors or weight files in step 109. The following uses… Figure 5 The example describes each step in more detail: (1) Record vibration data, in example ( Figure 5 The figure shows the time variation of the acceleration amplitude of the support load in the spatial direction x, as determined at the vibration isolator. (2) Generate spectrograms for use as network input variables here. In the example, each spectrogram represents a process to be evaluated as “good” or “bad”.

[0115] (3) Labeling, that is, assigning the spectrogram to the defined category "good" or "bad".

[0116] (4) Optionally, variations can be generated by enhancing the spectrogram. This can be done manually, but preferably with computer support.

[0117] (5) The network used requires spectrograms for learning, testing, and validation during the learning process. For this purpose, spectrograms are assigned to previously defined categories, but this assignment can also be done automatically by the program. There must be enough spectrograms available for each category (e.g., 1,000 or 10,000 per category).

[0118] (6) Create a network, in this example a CNN network. The network has two output neurons, corresponding to two defined categories ("good" and "bad").

[0119] (7) Perform the learning process.

[0120] (8) Generate a weight file containing parameterizations of each element and connection of the CNN network.

[0121] Figure 5 This diagram illustrates examples of the vibration behavior of an xy stage (“stage”) during different phases of operation within one cycle in three spatial directions. The upper graph shows the amplitude of the three vibration behaviors measured by the isolator accelerometer in the x, y, and z spatial directions over time during the stationary phase (before stage acceleration), followed by the acceleration measured by the sensor during stage acceleration, then the acceleration measured by the sensor during stage uniform motion, followed by deceleration, and finally the stationary phase after stage motion. The lower graph shows the acceleration of the stage in the x and y spatial directions over time. It is evident that in the example shown, acceleration or deceleration occurs only in the y direction, while vibration or acceleration is measured in all spatial directions.

[0122] In the example shown, it's easy to see that each individual operating phase or cycle is characterized by a specific frequency response. The characteristics of this frequency response (frequency, amplitude, time series) depend on whether the machine is still delivering the required performance data. Other information may also be available. For example, system faults affecting the frequency response, such as a damaged bearing, can be detected, provided that the bearing is impacting the frequency response. This damaged bearing can then lead to a "bad" classification, and with proper neural network design, it may even be possible to directly identify the problematic bearing as the source of the fault.

[0123] For training purposes, a sufficient number of spectrograms are generated, each corresponding to a complete run through four operational phases based on a full stage cycle. These are "labeled," i.e., classified into a specified category ("good" / "bad").

[0124] Figure 6 The spectrum diagram shown is for illustrative purposes only. This spectrum diagram may contain vibration data of the xy stage in one of the x, y, or z spatial directions. However, vibrations in the x, y, and z spatial directions can also be superimposed on a single spectrum diagram, as shown below. Figure 6 As the example shows, although Figure 6 Only a spectrum diagram in one spatial direction is displayed.

[0125] Vibrations in the x, y, and z spatial directions can also be displayed in separate spectrograms and processed together in an artificial neural network.

[0126] Vibrational behavior along rotational axes in the x, y, or z directions can also be recorded and processed. This allows for classification using more information, which can be useful and improve the reliability of the classification. However, the additional information also increases the complexity of the neural network, so it makes sense to determine the information ultimately needed for "good" and "bad" classification in a specific application.

[0127] Vibration data can also be represented in other ways, such as in the time or frequency domain. This also applies to the embodiments described herein and below.

[0128] like Figure 6 The spectrogram shown represents a collection of numerous measurements. This spectrogram contains time curves of a large number of frequencies, i.e., the spectral composition of vibrations. In performance analysis, this large number of measurements, or the whole, is evaluated, and the result is represented as an output value. This can be a category label, such as indicating good or poor operating conditions. Not only in this specific example of a spectrogram, but also in general, measurements are evaluated holistically in a neural network, which typically involves complex weighting and calculations of the measurements to obtain the output value.

[0129] exist Figure 7 In the illustrated embodiment, an artificial neural network is trained to generate classifications into "good" or "bad" categories from vibration data in the spatial direction x. For illustration, three spectrograms from vibrations in the spatial direction x are shown in each case, classified into the "good" or "bad" category. It goes without saying that in practice, not only three spectrograms are used, but preferably multiples thereof, such as 100 or more, or even 1,000 or more.

[0130] For classification, the "weight file" or weight file 13 generated during training is loaded into network 15, and the spectrum or more general measurement data 14 obtained during the normal operation of the vibration isolation system 1 is input into network 15. Figure 8 The data is then categorized, assigned to at least two categories 16 and 17, preferably to the "good" or "bad" category.

[0131] exist Figure 9 and Figure 10 In the further second embodiment based on the foregoing embodiments shown, the motion phases of the xy stage—acceleration, (uniform) motion, deceleration, and stationary states—are distinguished. This allocation allows for a more accurate classification of the xy stage's operating states; that is, the classification can be performed more reliably because the vibration data for these motion phases are different, rather than based on... Figure 5 The embodiments assume that the motion is performed in a summation view of the complete motion cycle, which includes all four motion phases.

[0132] Assume the input data is the vibration of the vibration sensor of vibration isolation system 1 during the movement of the xy stage in the spatial direction (e.g., spatial direction x).

[0133] As a result, information on the setup or adjustment quality of the vibration isolation system 1 can be provided for each motion phase in the "good" and "bad" categories respectively.

[0134] This enables more accurate performance analysis of vibration behavior at each stage of motion, thereby allowing for even more accurate detection of deviations of the xy stage or more general support equipment or machinery from specifications, such as bearing damage discussed above.

[0135] Training utilizes a sufficient number of spectrograms 20, each corresponding to a specific motion phase (acceleration, motion, deceleration, stationary position) or a specific combination of motion phases, such as the phase from stationary position to reaching maximum speed. Spectrograms 20 represent motion variable data provided by sensors 6a-6d, containing the time progression of multiple vibration frequency amplitudes. These spectrograms 20 are also categorized into specified classes, in this example, "good" and "bad". Furthermore, as... Figure 9 As shown in the example, each spectrogram is also assigned information about the motion phase or the corresponding period of the xy stage. Based on the four motion phases of the xy stage, Figure 9 The allocation is as follows: - Data P1: Data from Phase 1 of the motion, acceleration - Data P2: Data from phase 2 of the movement, movement - Data P3: Data from phase 3 of motion, deceleration - Data P4: Data from phase 4 of motion, rest phase Then, training was performed using spectrograms 20 from multiple different motion phases, which were separated according to motion phase. Figure 10 .

[0136] For classification, the "weight file" or weight file 13 generated during training is loaded into the artificial neural network, and measurement data 14 from normal machine operation, particularly in spectrogram form and motion phase assignments, is input into the network 15. This data is then classified as "good" or "bad," and the output is assigned to the corresponding motion phase. Figure 11 .

[0137] According to a further third embodiment of the present invention, additional information is to be used for performance analysis. Otherwise, this embodiment is based on the foregoing embodiments.

[0138] In the third embodiment, additional information is added, or different types of information are combined. The input data may include, for example, the following information, although the following list should not be considered exhaustive: - Vibration information of vibration isolation system 1 and / or supporting load 4 (especially the supported machine or equipment). - Vibration information of connecting components (such as vacuum pumps) - Vibration information from the floor or mounting surface of the vibration isolation system 1, or vibration information from components in the machine, such as vibration at the optical sensor, the electron beam tube, or the vacuum pump. - Intensity of acoustic vibration - Temperature of the environment, vibration isolation system 1 and / or supporting load 4 (especially the supported machine or equipment) - The intensity of electromagnetic radiation at the machine or equipment.

[0139] Vibration information also involves vibration behavior, including amplitude, period duration, frequency and / or frequency spectrum.

[0140] to this end, Figure 12 Showing sample selections of input data, where the following assignments apply: - X1: Vibration information of vibration isolation system 1 - X2: Vibration information of the vacuum pump - Xn: Measurement values ​​of temperature, electromagnetic radiation, and acoustic vibration. As a result, when the input data is appropriately combined, the generated output data can provide information about the performance of the vibration isolation system 1 and / or the supporting load 4 (especially the supported machine or equipment), such as operating parameters indicating whether the vibration isolation system 1 and / or the supporting load 4 (especially the supported machine or equipment) are performing well or poorly. In this manner, information about performance data can also be obtained independently of vibration information. For example, if a corresponding network is trained, electromagnetic radiation (characterized by radiation frequency and / or radiation amplitude and / or radiation direction) can also be used to classify machine performance data.

[0141] Similarly, training requires a sufficient amount of information, which, after proper processing, is "labeled," that is, assigned to specific categories. Figure 13 .

[0142] Figure 14 This shows an example of a process for classifying data based on measurements or input data obtained during normal operation.

[0143] According to a further fourth embodiment of the present invention, performance analysis of the vibration isolation system 1 is used to detect the cause of the failure. Otherwise, this embodiment is based on the foregoing embodiments.

[0144] As a result, when the input data is combined with the output data, the generated output data can provide information about whether the vibration isolation system 1 and / or the supporting load 4 (especially the supported machine or equipment) are operating within the specified specifications or deviating from them.

[0145] It can generate information about whether the vibration isolation system 1, or the machine or equipment, is operating within specifications. If it is operating outside specifications, it can determine which component is the cause. This assumes that the input data or measurements taken during normal operation provide sufficient information to achieve the appropriate classification.

[0146] Input data may include, for example, information indicating faulty or poorly adjusted components or environmental conditions. The following list is illustrative and should not be considered exhaustive: - Vibration information of vibration isolation systems (especially machines or equipment) operating within specifications. - Vibration information of vacuum pumps operating within specifications - Vibration information from vibration isolation systems (especially machines or equipment) that are not operating within specifications. Vibration information of vacuum pumps operating outside of specifications Vibration information also involves vibration behavior, including amplitude, period duration, frequency and / or frequency spectrum.

[0147] to this end, Figure 15 The selection of input data is shown only as an example, where the following assignments apply. - X1: Vibration information of vibration isolation system 1 - X2: Vibration information of the vacuum pump Then, instruction is performed using multiple input datasets representing various possible combinations. For the example above, the following input data could be used: Vibration isolation system 1 operates within specifications, and the vacuum pump operates within specifications. - Vibration isolation system 1 is not operating within specifications, while the vacuum pump is operating within specifications. - Vibration isolation system 1 is operating within specifications, while the vacuum pump is not operating within specifications. In addition to the information already discussed, the input data therefore includes information indicating faulty components or environmental conditions. Further input data may also relate to the supporting load 4, particularly the supported machine or equipment, such as an XY stage.

[0148] For example, the frequency spectrum generated when the xy stage moves correctly in the y direction can be used for teaching. If the bearings used when the xy stage moves in the y direction fail, a deviation in the frequency spectrum may occur during normal operation. Frequency spectrum deviations can also have various causes, such as component aging, faulty / poor repairs or software updates, unoptimized changes to motion sequences within the machine, poor cable connections, and compressed air fluctuations. Data from such components can also be used as input variables.

[0149] The "weight file" is trained and created as described above, but it is essential to ensure that there is a sufficiently large amount of data records for each category. Figure 16 .

[0150] Figure 17 This shows an example of a process for classifying data based on data obtained during regular operation.

[0151] To perform classification, a "weight file" generated during the training of the artificial neural network is loaded into the network, along with the machine operation data to be classified. This data is then classified and assigned to faulty components where applicable.

[0152] In the above embodiments, the sensor is not only arranged in or on the vibration isolation system 1, but also on the supporting load 4, particularly on the plate or the machine supported thereon, or arranged in or on the supporting equipment.

[0153] In addition, sensors may also be disposed on or within associated components, such as on the xy stage, optical sensors, optical or electro-optical elements of a microscope, on spectral, scattered light, or interferometric measuring equipment, or on a scanning probe microscope, on a distance sensor (e.g., for measuring the distance between an optical element and a substrate), on a cover plate, on a vacuum pump, or on other components of a vibration isolation system. Sensors may include, for example, optical sensors, ultrasonic sensors, accelerometers, or distance sensors; in this example, an accelerometer is provided.

[0154] Further sensors can be assigned to the environment of the vibration isolation system 1, such as the floor or mounting surface, or to the surrounding environment, such as other components near the vibration isolation system. In particular, it is recommended to select components that can transmit vibrations to the vibration isolation system and / or support loads. This could be, for example, a ceiling or nearby overhead crane, such as a workshop crane or bridge crane.

[0155] To detect electromagnetic fields, a magnetic field sensor, fluxgate magnetometer, or saturated iron core magnetometer is provided.

[0156] A pressure sensor is provided to detect the intensity of the sound field.

[0157] Thermocouples are provided for measuring the temperature at the installation site and / or at vibration isolation system components and / or at supporting loads.

[0158] The generation of output data from input data and the classification for performance analysis are performed computer-aided, and the program can be stored in computing device 3. Of course, the program or components of the artificial neural network can also be stored in other locations.

[0159] For example, in a further embodiment of the invention, it is envisioned that the components of the artificial neural network 10 are centrally stored in a higher-level computing device and used in multiple distributed vibration isolation systems 1. This has advantages in terms of maintenance and training.

[0160] In a further embodiment of the invention, the artificial neural network may also be designed to generate control parameters for the control and regulation of one or more vibration isolation systems 1 connected to a higher-level computing device, using information or input data from multiple vibration isolation systems 1 after performance analysis.

[0161] Variations can be generated by enhancing spectrograms or other forms of vibration representation (i.e., by altering or transforming the dataset), which means that it is preferable to also use a computer to introduce deliberate biases in the spectrograms or input variables to increase the number of different training datasets.

[0162] This is advantageous because training requires a large amount of data, among other things, to capture the necessary dispersion of input information that occurs during regular operation. Such data can also be artificially or computer-assistedly generated within certain limits. For example, noise can be added to vibration data.

[0163] In this way, by recording a large number of localization sequences and possible augmentations, a large number of spectrograms can be generated for training the artificial neural network used. For this purpose, the spectrograms are assigned to desired categories or classifications. According to one embodiment of the invention, this assignment can also be performed automatically by computer, which significantly simplifies manual data preparation.

[0164] For effective performance analysis and diagnostics, it is helpful to have a sufficient number of spectrograms available for the relevant category. In some cases, 10 to 100 input data points or spectrograms per category may be sufficient, but it is best to provide, for example, approximately 1,000 or even 10,000 or more datasets to improve the quality of the performance analysis.

[0165] During the normal operation of the vibration isolation system 1, a pre-trained or learned artificial neural network is used. In an advantageous embodiment of the invention, the trained artificial neural network is further trained during operation.

[0166] In this way, vibration isolation system 1 can be easily adapted to the conditions commonly found at installation sites.

[0167] In a further advantageous embodiment of the invention, the artificial neural network uses training data from a previous learning process or from another external source.

[0168] Performance analysis can be used to generate information or output data, such as control parameters for controlling or adjusting the vibration isolation system 1 and / or the support load 4. Therefore, an important aspect of the present invention is that these control parameters can be used directly or indirectly to control or adjust the vibration isolation system 1 and / or the support load 4.

[0169] Therefore, the present invention also includes a method for controlling or regulating the vibration isolation system 1, wherein control parameters are obtained during operation as part of the performance analysis described above and used to control the vibration isolation system 1. This provides a self-learning vibration isolation system 1 that can, for example, respond to environmental changes (such as vibrations occurring through the floor).

[0170] The vibration isolation system 1 is designed as an active, fixed vibration isolation system. In this embodiment, it includes four vibration isolators that can actively counteract vibrations that occur during normal operation via four actuators.

[0171] Another aspect of the present invention includes a method for diagnosing, and in particular analyzing, the performance of a vibration isolation system 1 including the vibration isolation system described above.

[0172] This method can be used to perform performance analysis on vibration isolation systems and also on supporting loads.

[0173] A further aspect of the invention provides adjustments to the artificial neural network. This may become necessary if influencing factors change during network use, for example: - More data may be generated during network usage, which can be used to improve classification. For example, due to software improvements, the installed machine may be better able to cope with interference. - Changes in the products being processed by machines place different demands on them. Since retraining artificial neural networks can be costly, this further approach suggests continuous optimization during the regular operation of the artificial neural network. This saves retraining time and shortens the interval between network updates. Figure 18 The process is illustrated schematically in the diagram. Here, while applying the AI ​​model, the AI ​​model is continuously trained using external performance and evaluation metrics that classify the system state. The AI ​​model is understood here as a model containing the neural network described above.

[0174] Besides the architecture described above for classifying input data, other methods exist. In the example above, the input data is evaluated over a time period, such as a sequence consisting of acceleration, motion, deceleration, and stationary phases over two seconds.

[0175] Architectures also exist for processing time-series data. One example is the LSTM ("Long Short-Term Memory") neural network, which can process sequences of multiple smaller time periods in an additional dimension. This architecture can be similarly used for the classification tasks mentioned above.

[0176] List of reference numerals 1 Vibration isolation system 2a-2d Vibration isolators 3. Computing equipment 4. Load 5a-5d actuator 6a-6d sensors Spectrum Atlases 10 and 11 13 Weighted Files 14 Measurement Data 15 Network Categories 16 and 17 20. Spectrum Diagram 101 Steps: Data Acquisition 102 Steps: Data Processing 103 Steps: Labeling 104 steps: Enhancement 105 Training Spectrum 106 Test Spectrum 107 Steps: Network Creation 108 Steps: Network Training Step 109: Create a weight vector or weight file.

Claims

1. A vibration isolation system, particularly having diagnostic capabilities, preferably a fixed active vibration isolation system, capable of supporting vibration isolation loads, comprising: At least one vibration isolator, preferably multiple vibration isolators, capable of counteracting vibrations occurring during operation by means of an actuator. At least one sensor, preferably multiple sensors, is designed to record measurements related to the vibration isolation system and / or supporting load and / or environment, and to provide these measurements as input data to a computer-aided method for performance analysis of the vibration isolation system and / or supporting load.

2. The vibration isolation system according to the preceding claims, wherein the computer-aided method comprises or is based on at least one artificial neural network.

3. A vibration isolation system (1), particularly a vibration isolation system according to any one of the preceding claims, for supporting a load in a vibration-isolated manner, comprising: The system comprises at least one, preferably multiple, vibration isolators (2a-2d) and at least one sensor (6a-6d) and a computing device (3), wherein the sensor (6a-6d) is designed to detect motion variables, particularly mechanical vibration, motion or acceleration, and wherein the computing device (3) includes an artificial neural network (10), wherein the artificial neural network (10) is designed to process motion variable data provided by the at least one sensor (6a-6d), classify whether the vibration isolation system (1) and the supporting load are in an operating state with sufficiently small or excessive motion disturbance, and output the classification.

4. The vibration isolation system according to any one of the preceding claims, characterized in that, The motion variable data provided by the at least one sensor (6a-6d) includes a spectrogram (20) which contains the time progression of the amplitude of multiple vibration frequencies.

5. The vibration isolation system according to any one of the preceding claims, wherein a trained or learned artificial neural network is used in normal operation.

6. The vibration isolation system according to any one of the preceding claims, wherein the supporting load is supported by three or four vibration isolators.

7. The vibration isolation system according to any one of the preceding claims, wherein the supporting load includes processing machinery or components thereof, particularly in the semiconductor industry, such as machines for processing semiconductor devices and / or nanostructure substrates or elements, particularly machines for processing wafers, masks or displays, lithography equipment, or measuring equipment or devices, particularly measuring equipment or devices for measuring semiconductor elements and / or nanostructure substrates, such as electron microscopes, 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.

8. The vibration isolation system according to any one of the preceding claims, wherein the input data includes the vibration behavior of the vibration isolation system, particularly amplitude, period duration, frequency, or acceleration.

9. The vibration isolation system according to any one of the preceding claims, wherein the input data includes the temperature at the installation location and / or at a component of the vibration isolation system and / or at a supporting load.

10. The vibration isolation system according to any one of the preceding claims, wherein the input data includes the intensity of the electromagnetic field at the installation location.

11. The vibration isolation system according to any one of the preceding claims, wherein the input data includes the intensity of the sound field at the installation location.

12. The vibration isolation system according to any one of the preceding claims, wherein the sensor is arranged in or on the vibration isolation system, and / or on the support load, and / or in the environment, particularly on the ground, on the mounting surface, or on a component or element near the vibration isolation system.

13. The vibration isolation system according to any one of the preceding claims, wherein the artificial neural network forms at least one model comprising at least one convolutional level (convolutional layer) and / or at least one neuron level (neural layer), and / or other components, such as activation layers.

14. The vibration isolation system according to any one of the preceding claims, wherein the artificial neural network can be trained, particularly during operation of the vibration isolation system, and / or using training data from a previous learning process.

15. The vibration isolation system according to any one of the preceding claims, wherein the artificial neural network uses at least one method of regression, machine learning or deep learning.

16. The vibration isolation system according to any one of the preceding claims, wherein, for training the artificial neural network, the dataset is computer-aidedly modified or transformed to obtain multiple training datasets.

17. The vibration isolation system according to any one of the preceding claims, wherein control parameters for controlling the vibration isolation system and / or the support load are generated based on the performance analysis.

18. The vibration isolation system according to any one of the preceding claims, wherein control parameters are used to control the vibration isolation system and / or the support load.

19. A method for controlling or regulating a vibration isolation system, preferably a vibration isolation system according to any one of the preceding claims, wherein control parameters for controlling the vibration isolation system are obtained and used during operation by means of performance analysis.

20. A method for diagnosing, and particularly for performance analysis, a vibration isolation system, preferably a vibration isolation system according to any one of claims 1 to 14.

Images