METHOD AND DEVICES FOR MONITORING THE CONDITION OF A DENTISTRY MACHINE

Description

TECHNICAL AREA

[0001] The present invention relates to methods for monitoring the condition of a gear cutting machine for machining toothed workpieces, in particular a gear grinding machine. Methods according to the preambles of claims 1, 6 and 10 are known from document US 2014 / 0256223 A1.

[0002] The invention also relates to a method for creating a training dataset for a machine learning algorithm according to the preamble of claim 8 and a method for training a machine learning algorithm according to the preamble of claim 9. Methods according to the preambles of these claims are known from document US 2018 / 0264613 A1.

[0003] The invention also relates to devices for monitoring the condition of a gear cutting machine with a plurality of machine axes according to the preambles of claims 14 and 15. Devices according to the preambles of these claims are known from document CH 715989 A1. STATE OF THE ART

[0004] During the hard finishing of toothed workpieces, especially external or internal gears, in a gear cutting machine, manufacturing deviations naturally occur. These deviations manifest as discrepancies between the actual manufactured geometry of the workpieces and a predetermined target geometry. These manufacturing deviations can be caused, among other things, by malfunctions or wear of the various components of the gear cutting machine, or by unsuitable assembly of the components. For example, a manufacturing deviation can be caused by a drive moving a slide of the gear cutting machine to a different position than the target position specified by the machine control, by a worn spindle bearing, or by machine parts being connected in an unsuitable manner, resulting in insufficient vibration damping.

[0005] Hard finishing is usually the final step in workpiece machining. After this step, the workpieces are installed in gearboxes. Following assembly, the gearboxes undergo a final inspection ("EOL inspection") on an EOL test bench (EOL = End of Line). This inspection focuses particularly on the gearbox's noise characteristics. Manufacturing deviations in the workpieces often lead to undesirable noise. Even the smallest manufacturing deviations, such as those that can occur during machining on a perfectly functioning machine, can cause noise. Especially in electric drives, the gearbox is sometimes the dominant noise source, and any noise development within the gearbox is particularly disruptive.

[0006] It is desirable to be able to predict the noise behavior caused by a workpiece before the workpiece is installed in a gearbox, in order to avoid costly disassembly required if the gearbox proves to be prone to noise during end-of-life (EOL) testing. It is also desirable to be able to draw direct conclusions about specific causes within the gear cutting machine from the measured noise behavior of a gearbox.

[0007] US20140256223A1 discloses a method for hard finishing of gear teeth with corrections and / or modifications on a gear cutting machine, wherein gear pairs that mesh with each other within a gearbox or testing device are machined taking into account their respective mating flanks, and wherein the tooth flanks of the affected workpieces are provided with periodic waviness corrections or modifications. The rotational error profile is determined by measuring the rotational path error of the gear pairs in a gear measuring device and / or in a gearbox. This measurement result serves as input for defining the amplitude, frequency, and phase for the periodic flank waviness corrections on the tooth flanks of the gear pairs for manufacturing in the gear cutting machine. PRESENTATION OF THE INVENTION First aspect: Prediction of noise excitations

[0008] In a first aspect, it is an object of the present invention to provide a method for monitoring the condition of a gear cutting machine, with which the development of disturbing noises in a gearbox containing a workpiece machined with the gear cutting machine can be predicted at an early stage.

[0009] This problem is solved by a method according to claim 1. Further embodiments are specified in the dependent claims.

[0010] A method for monitoring the condition of a gear cutting machine with a plurality of machine axes is therefore specified, which comprises the following steps: a) Performing a test cycle in which at least some of the machine axes are systematically actuated and associated machine measurement data are determined; b) Performing a spectral analysis of the machine measurement data, in which machine spectral data are calculated from the machine measurement data; and c) Determining predicted EOL spectral data based on the machine spectral data, wherein the predicted EOL spectral data indicate at which orders excitations are to be expected in an EOL spectrum when a workpiece machined with the gear cutting machine is installed in a gearbox and rolls on a mating gear in the gearbox.

[0011] The procedure may include: d) Outputting the predicted EOL spectral data or at least one derived quantity.

[0012] Within the proposed procedure, machine measurement data is acquired during a test cycle while specific machine axes are actuated. The test cycle takes place during a machining break of the gear cutting machine; that is, the machining tool of the gear cutting machine is not engaged with a workpiece during the test cycle. The machine measurement data can include, in particular, acceleration values ​​determined by an accelerometer, position values ​​determined by a position sensor, and / or current values ​​determined by a current sensor. From the machine measurement data, machine spectral data (spectral state data for the machine) are calculated using spectral analysis (especially order analysis) to determine at which frequencies or orders periodic excitations of machine components occur while a machine axis is actuated.

[0013] End-of-Life (EOL) spectral data are predicted based on machine spectral data. This prediction is based on the following considerations: Excitations from the machine components are transferred to the workpieces during machining. If the machining tool is dressed using the machine axes, these excitations are also transferred to the dressed tool, from where they are further transferred to the workpiece during machining. Based on the known kinematic relationships between the individual machine components and taking into account the selected machining parameters, it is possible to calculate how such excitations affect the manufactured workpieces. In particular, such excitations can lead to periodic deviations (waviness) on the tooth flanks of the workpieces.The term "kinematic link" refers to the way in which the movement of one component is transferred to another during the machining process. The kinematic link between the tool spindle and the workpiece spindle plays a central role in the machining process; that is, it describes how the movements of the tool correlate with the movements of the workpiece. If the gear cutting machine is a machine for the rolling machining of workpieces, such as a gear grinding or skiving machine, this link is characterized by the rolling engagement between the workpiece and the tool and is determined by the workpiece geometry and the tool geometry. The way in which a periodic excitation of the tool is transferred to the workpiece flanks as a waviness through this link can be easily calculated.In particular, it is easy to calculate the order, relative to the workpiece rotation, of an excitation in the gearbox caused by waviness, if the order of the resulting excitation of the tool, relative to the tool rotation, is known. The ratio between these two orders is referred to below as the "transmission factor." It is also possible to calculate how vibrations of one component affect the other for all other components between which motion is transmitted, and in particular, the corresponding transmission factors between the disturbance orders of these components and the resulting disturbance orders in the end-of-line (EOL) spectrum can be calculated.

[0014] The calculation of the predicted EOL spectral data can therefore include the application of a transfer factor to the machine spectral data, where the transfer factor depends on the kinematic link between the machine axis for which the machine spectral data were determined and the workpiece.

[0015] The described method yields predicted end-of-line (EOL) spectral data. This predicted EOL spectral data is based on an examination of the actual machine condition and therefore incorporates sources of interference within the machine that may have been unknown a priori and thus cannot be included in a calculation or simulation of EOL spectral data based solely on the known machine structure. This allows the method to reliably predict at which order variations in the finished gearbox noise excitation will occur. In this way, noise excitation can be predicted even before the workpieces are actually installed in the gearbox. The affected workpieces can be removed and, if necessary, examined more closely, thus preventing the costly dismantling of fully assembled gearboxes.Furthermore, measures can be taken to identify and eliminate the source of the error that leads to the expected noise developments.

[0016] It is important to note that determining the predicted EOL spectral data does not necessarily involve a quantitative prediction of noise intensities at the different orders, but rather a qualitative indication of which orders are actually "disturbance orders," i.e., at which orders a significant noise intensity can be expected. The predicted EOL spectral data may include, in particular, the disturbance orders and associated intensity indicators, where the intensity indicators represent (possibly only very rough) estimates of expected disturbance intensities at the disturbance orders.

[0017] The predicted end-of-line (EOL) spectral data can be determined individually for each actuated machine axis. This means that separate machine measurement data is acquired for each machine axis activated during a test cycle. Separate machine spectral data is then calculated from this separate measurement data, and based on this, separate EOL spectral data is predicted for each actuated machine axis. This approach makes it possible to predict which machine axis might cause which disturbance orders in the EOL spectral data.

[0018] Preferably, steps a) to c) are repeated multiple times, with workpieces being machined on the gear cutting machine between test cycles and the test cycles being performed during machining breaks when the machining tool is not engaged with a workpiece. The development of the predicted EOL spectral data as a function of the test cycle or time is then visualized and / or analyzed. This approach is based on the consideration that the predicted EOL spectral data may sometimes be of limited value based on a single test cycle. However, wear or failure of components of the gear cutting machine can occur during machining, which then manifests itself in a significant change in the predicted EOL spectral data. Therefore, it is proposed to consider the temporal development of the predicted EOL spectral data.By appropriately visualizing the temporal changes of the predicted end-of-line (EOL) spectral data, disturbance orders where significant changes in noise excitation behavior are expected over time can be easily identified. The evolution of the calculated EOL spectral data can also be analyzed numerically. For example, a numerical analysis could involve performing a regression analysis of the expected disturbance intensities for selected or all disturbance orders using suitable regression functions, such as a polynomial of at least second order. Based on this analysis, for instance, a warning indicator can be determined and issued if the analysis shows that a gradient in disturbance intensity is expected for at least one disturbance order that meets a specific warning criterion.

[0019] If reference machine spectral data are available for many reference machines, acquired through numerous different reference test cycles, reference EOL spectral data can be predicted from this data. It is then possible to automatically evaluate the predicted EOL spectral data of the machine under assessment by comparing it with the predicted reference EOL spectral data of the reference machines or with derived parameters. This allows for the automatic identification of expected noise problems, without requiring specialized knowledge or measured EOL spectra as a basis for evaluation. In particular, a statistical analysis of the predicted reference EOL spectral data can be performed for this purpose.Regarding the considerations underlying this approach and further implementation possibilities, reference is made to the patent application filed on the same day as the present application by the same applicant entitled "Method for monitoring the condition of a machine tool", the content of which is fully incorporated into the present disclosure by reference. Second aspect: Identification of components that cause noise problems

[0020] The reverse approach is also possible: measuring end-of-line (EOL) values ​​on the EOL test bench while the workpiece rolls on a mating gear in the gearbox, performing a spectral analysis of the EOL measurements to determine the measured EOL spectral data, and drawing conclusions from this EOL spectral data about individual components of the gear cutting machine whose condition causes disturbances in the EOL spectral data. The EOL measurements can be determined by any suitable sensors on the EOL test bench, in particular accelerometers and sensors for detecting rotational errors.

[0021] A method according to claim 6 for monitoring the state of a gear cutting machine with a plurality of machine axes is therefore specified, comprising the following steps: a) Performing an EOL test on a gearbox containing a workpiece machined by the gear cutting machine, wherein in the EOL test the workpiece is rolled on a mating gear in the gearbox and associated EOL measurement data are determined; b) Performing a spectral analysis of the EOL measurement data, wherein EOL spectral data are calculated from the EOL measurement data; and c) Determining predicted condition data based on the EOL spectral data, wherein the predicted condition data for at least one machine axis indicate which orders of this machine axis are consistent with the calculated EOL spectral data.

[0022] The procedure may include: d) outputting the predicted condition data or at least one derived quantity.

[0023] This method allows conclusions to be drawn about which machine axes, and possibly which components of these machine axes, are responsible for noise that actually occurs after a workpiece machined with the gear cutting machine has been installed in a gearbox. This method also utilizes the fact that the orders of noise, relative to the rotation of the workpiece in the gearbox, can be readily calculated from the orders of measurement data determined for the components of the gear cutting machine, given knowledge of the kinematic relationships between the components and taking into account the selected machining parameters.

[0024] In particular, this method can be performed without requiring condition measurements on the gear cutting machine itself. However, significant advantages arise when this method is combined with condition measurements on the gear cutting machine. Therefore, the method can additionally include: e) Performing a test cycle in which at least some of the machine axes are systematically actuated and associated machine measurement data are determined; f) Performing a spectral analysis in which machine spectral data are calculated from the machine measurement data; and g) Determining predicted EOL spectral data based on the machine spectral data, wherein the predicted EOL spectral data indicate at which orders excitations are to be expected in an EOL spectrum when a workpiece machined with the gear cutting machine is installed in a gearbox and rolls on a mating gear in the gearbox. where determining the predicted condition data involves comparing the EOL spectral data calculated from the EOL measurement data with the predicted EOL spectral data.

[0025] By comparing EOL spectral data obtained through measurements during a gearbox test run with EOL spectral data predicted from measurement data of the gear cutting machine, the causes of noise disturbances can be determined particularly reliably. Third aspect: Use of a machine learning algorithm

[0026] The methods discussed so far require knowledge of the kinematic links between components of the gear cutting machine. In a further aspect, the invention provides a method according to claim 10 that makes it possible to predict the noise behavior of a gearbox based on measurements of the machine's condition, or to draw conclusions about the condition of the gear cutting machine from the measured noise behavior of a gearbox, even without knowledge of the kinematic links. This method employs a trained machine learning algorithm whose input variables are condition data of the gear cutting machine and whose output variables are predicted end-of-line (EOL) data that characterize the expected noise behavior of the gearbox, or whose input variables are EOL data and whose output variables are predicted condition data that characterize an expected state of the machine.

[0027] The machine learning algorithm is trained using the following method according to claim 9: a) Performing a test cycle in which at least some of the machine axes are systematically actuated and associated state data are determined by measurements; b) Machining at least one workpiece with the gear cutting machine while the gear cutting machine is in a state characterized by the state data; c) Installing the machined workpiece in a gearbox; d) Performing an EOL test on the gearbox, in which the workpiece is rolled on a mating gear in the gearbox and associated EOL data are determined; e) Storing the state data and the corresponding EOL data in a training dataset; f) Repeating steps a) to e) for a large number of test cycles and machined workpieces, wherein the workpieces have the same target geometry and are machined under the same machining conditions; and g) Training the machine learning algorithm with the training dataset.

[0028] The training dataset therefore contains a large number of state data with the corresponding EOL data for a large number of workpieces that have the same target geometry, were machined under the same machining conditions and were installed in the same type of gearbox.

[0029] The target geometry includes, in particular, parameters such as the normal module, number of teeth, and helix angle of the gear teeth, but can also include further parameters such as specified tooth flank modifications. Machining conditions are considered to be identical, in particular, when the machine axes move in the same way during the machining operations. For example, if gear grinding is used as the machining process, the machining conditions are identical when the workpieces are machined with the same radial infeed, the same axial feed rate, and the same shift speed, when the tool speed is the same for all workpieces, and when the grinding worm used has the same number of threads and the same pitch height for all workpieces, so that the associated workpiece speed is also the same.If the grinding screw is a dressable grinding screw that is dressed with a rotating sliding dressing tool, the dressing conditions are also part of the machining parameters, in particular the speed of the tool spindle and the speed of the dressing tool during the dressing process.

[0030] The machine learning algorithm is trained with the state data and the corresponding end-of-life (EOL) data. As a result, the machine learning algorithm can predict EOL dates based on state data, or vice versa, without requiring knowledge of the kinematic connections between the components of the gear cutting machine.

[0031] Many different types of machine learning algorithms are known that can be used in this context, and the structure of the training dataset can vary accordingly. Classification algorithms are particularly suitable for practical implementation. For this purpose, the output variables can be reduced to a limited number of classes. For example, if the input variables are end-of-line (EOL) data and the output variables are predicted state data, the predicted state data could consist of, for example, a single real value per machine axis. Each value could then, for example, indicate a probability that the respective machine axis is responsible for the observed EOL data. The training data should then contain state data representing a single real state value per machine axis and the corresponding EOL data.If the input variables are state data and the output variables are predicted end-of-life (EOL) data, the predicted EOL data could, for example, consist of one real value per order for a relatively small number of orders (particularly important in practice). Each value could then, for example, represent a predicted relative spectral intensity of the respective order. The training data should then contain corresponding EOL data. Of course, entirely different, even more complex, output variables are also conceivable. For practical implementation, an artificial neural network (ANN) or a support vector machine (SVM) is suitable, for example. In a particularly simple example, the input variables could be state data, and the output variable could be a single real value that characterizes the overall noise behavior of the entire gearbox on the EOL test bench.A Random Forest, for example, is a suitable machine learning algorithm for predicting such a value. With such a value, an expected problematic noise behavior can be easily identified, and measures can be taken to prevent affected workpieces from being installed in gearboxes.

[0032] The condition data can generally include various types of data that correlate with the state of a machine axis with respect to its vibration behavior. In particular, the condition data can include machine spectral data as defined in the context of the first and second aspects.

[0033] The EOL data can also include various types of data that correlate with the noise behavior of the gearbox. In particular, the EOL data can include EOL spectral data, as defined in the context of the first and second aspects.

[0034] The training data can be stored in a database. This database can be located remotely from the machine being monitored. It can also be implemented in the cloud, i.e., as a service-based computing resource shared by multiple users. An evaluation computer can access the database to train the machine learning algorithm. This evaluation computer is also preferably located separately from the machine tool. It is connected to the machine tool via a network connection. The evaluation computer does not need to be a single physical unit; it can also be implemented in the cloud.

[0035] The invention further provides a device according to claim 14 and claim 15 for monitoring the state of a gear cutting machine with a plurality of machine axes, comprising a processor and a storage medium on which a computer program is stored. When executed on the processor, the computer program causes at least a part of the process steps of one of the methods described above to be carried out. The invention also provides a corresponding computer program. The computer program can be stored on a non-volatile storage medium. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Preferred embodiments of the invention are described below with reference to the drawings, which serve only for illustration and are not to be interpreted restrictively. The drawings show: Fig. 1 a schematic view of a gear grinding machine; Fig. 2 a diagram illustrating spectral data acquired in measurement cycles; Fig. 3 a sketch of an EOL test bench; Figs. 4A and 4B excerpts from a table for converting orders of machine components into orders of a workpiece installed in a gearbox on an EOL test bench; Fig. 5 a diagram illustrating spectral prediction data; Fig. 6 a schematic representation of the temporal evolution of disturbance intensity at a disturbance order; Fig. 7 a sketch illustrating a machine learning algorithm; and Fig. 8 an excerpt of exemplary training data for the machine learning algorithm. DESCRIPTION OF PREFERRED EXECUTION FORMS Example of a gear grinding machine setup

[0037] In the Fig. 1As an example of a gear cutting machine, a gear grinding machine 1 is shown, which will also be referred to as "machine" in the following. The machine 1 has a machine bed 11 on which a tool carrier 12 is slidably guided along a radial feed direction X. The tool carrier 12 carries an axial slide 13, which is slidably guided along a feed direction Z relative to the tool carrier 12. A grinding head 14 is mounted on the axial slide 13, which can be pivoted about a pivot axis (the so-called A-axis) parallel to the X-direction to adapt to the helix angle of the gear being machined. The grinding head 14, in turn, carries a shift slide on which a tool spindle 15 can be slid along a shift direction Y relative to the grinding head 14. A helical grinding wheel (grinding worm) 16 is mounted on the tool spindle 15.The grinding screw 16 is driven by the tool spindle 15 to rotate about a tool axis B.

[0038] The machine bed 11 further supports a swiveling workpiece carrier 20 in the form of a rotary tower, which can be pivoted about a pivot axis C3 between at least three positions. Two identical workpiece spindles are mounted diametrically opposite each other on the workpiece carrier 20, one of which is located in the Fig. 1 Only one workpiece spindle 21 with its associated tailstock 22 is visible. A workpiece can be clamped onto each of the workpiece spindles and driven to rotate around a workpiece axis C1 or C2. The [unclear text] Fig. 1 The visible workpiece spindle 21 is in a machining position in which a workpiece 23 clamped on it can be machined with the grinding screw 16. The other, arranged offset by 180° and in the Fig. 1The workpiece spindle, which is not visible, is in a workpiece change position, in which a finished workpiece can be removed from this spindle and a new blank can be clamped on. A dressing unit 30 is mounted offset by 90° to the workpiece spindles.

[0039] Machine 1 thus has a large number of moving components, such as slides or spindles, which can be moved by corresponding drives. In technical circles, these drives are often referred to as "NC axes," "machine axes," or simply "axes." Sometimes this term also includes the components driven by the drives, such as slides or spindles.

[0040] Machine 1 also features a large number of sensors. Examples include: Fig. 1Only two sensors, 18 and 19, are shown schematically. Sensor 18 is an acceleration sensor (vibration sensor) for detecting vibrations of the grinding spindle housing 15. Sensor 19 is a position sensor for detecting the position of the axial slide 13 relative to the tool carrier 12 along the Z-axis. In addition, machine 1 includes a large number of other sensors. These include, in particular, further position sensors for detecting the actual position of each linear axis, rotary angle sensors for detecting the rotational position of each rotary axis, current sensors for detecting the drive current of each axis, and further vibration sensors for detecting vibrations of each driven component.

[0041] All driven axes of machine 1 are digitally controlled by a machine control system 40. The machine control system 40 comprises several axis modules 41, a control computer 42, and an operator panel 43. The control computer 42 receives operator commands from the operator panel 43 as well as sensor signals from various sensors of machine 1 and calculates control commands for the axis modules 41. It also outputs operating parameters to the operator panel 43 for display. The axis modules 41 provide control signals for one machine axis each at their outputs.

[0042] A monitoring device 44 is connected to the control computer 42.

[0043] The monitoring device 44 can be a separate hardware unit assigned to machine 1. It can be connected to the control computer 42 via a known interface, e.g., via the known Profinet standard, or via a network, e.g., via the Internet. It can be physically part of machine 1, or it can be located physically separate from machine 1.

[0044] During machine operation, the monitoring device 44 receives a variety of different measurement data from the control computer 42. Among the measurement data received from the control computer are sensor data acquired directly by the control computer 42 and data read by the control computer 42 from the axis modules 41, e.g., data describing the target positions of the various machine axes and the target current consumption in the axis modules.

[0045] The monitoring device 44 can optionally have its own analog and / or digital sensor inputs to directly receive sensor data from other sensors as measurement data. These other sensors are typically not directly required for controlling the actual machining process, e.g., accelerometers to detect vibrations, or temperature sensors.

[0046] The monitoring device 44 can alternatively be implemented as a software component of the machine control 40, which is executed, for example, on a processor of the control computer 42, or it can be designed as a software component of the service server 45 described in more detail below. The service server 45 has a processor 451 (shown only schematically) and a memory device 452.

[0047] The monitoring device 44 communicates directly or via the internet and a web server 47 with the service server 45. The service server 45, in turn, communicates with a database server 46 and its database DB. These servers can be located remotely from machine 1. The servers do not need to be a single physical unit. In particular, the servers can be implemented as virtual units in the so-called "cloud".

[0048] The service server 45 communicates with an end device 48 via the web server 47. The end device 48 can, in particular, run a web browser, which is used to visualize the received data and its analysis. The end device does not need to meet any special computing power requirements. For example, the end device could be a desktop computer, a notebook computer, a tablet computer, a mobile phone, etc. Machining of a batch of workpieces

[0049] The following describes how to process workpieces with machine 1.

[0050] To machine a blank workpiece, the workpiece is clamped onto the workpiece spindle in the workpiece change position by an automatic workpiece changer. The workpiece change occurs concurrently with the machining of another workpiece on the other workpiece spindle, which is in the machining position. Once the new workpiece is clamped and the machining of the other workpiece is complete, the workpiece carrier 20 is swiveled 180° around the C3 axis, so that the spindle with the new workpiece moves into the machining position. Before and / or during the swiveling process, a centering operation is performed using the associated centering probe. For this purpose, the workpiece spindle 21 is rotated, and the position of the tooth gaps of the workpiece 23 is measured using the centering probe 24. The pitch angle is then determined based on this measurement.

[0051] When the workpiece spindle, which carries the workpiece 23 to be machined, has reached the machining position, the workpiece 23 is brought into contact with the grinding screw 16 without collision by moving the tool carrier 12 along the X-axis. The workpiece 23 is then machined by the grinding screw 16 in a rolling engagement. During machining, the workpiece is continuously advanced along the Z-axis with a constant radial X-axis feed. In addition, the tool spindle 15 is slowly and continuously shifted along the Y-axis to ensure that unused areas of the grinding screw 16 are continuously engaged during machining (so-called shift movement).

[0052] Simultaneously with the workpiece machining, the finished workpiece is removed from the other workpiece spindle, and another raw part is clamped onto this spindle.

[0053] When, after machining a certain number of workpieces, the grinding screw 16 has become so worn that it is too dull and / or its flank geometry is too inaccurate, it is dressed. For this purpose, the workpiece carrier 20 is swivelled by ±90° so that the dressing device 30 is positioned opposite the grinding screw 16. The grinding screw 16 is then dressed with the dressing tool 33. In this case, the dressing tool is a rotating dressing disc. Test cycle of the gear cutting machine

[0054] During processing breaks, the monitoring device 44, in conjunction with the machine control 42, performs a test cycle to check the condition of individual or all components of machine 1. During such a test cycle, a selected part of the machine axes or all machine axes are systematically actuated, and measurements are taken on the machine.

[0055] For example, each linearly displaceable component is moved along with its associated machine axis, and the component's instantaneous position is continuously determined using the aforementioned position sensors. From this, a positional deviation between the target position and the measured actual position is continuously calculated and transmitted to the monitoring device 44. The same process can also be performed for the rotaryally driven spindles, in which case angle sensors are used to determine positional deviations.

[0056] The vibration behavior of selected machine axes is also determined while the respective machine axis is activated. This is done using acceleration sensors (vibration sensors) connected to these components. The results of the vibration measurements are also transmitted to the monitoring device 44.

[0057] Furthermore, the power consumption of the drive motors of the machine axes is continuously determined while they are activated. Current sensors integrated into the axis modules 41 can be used for this purpose. The results of the current measurements are also transmitted to the monitoring device 44.

[0058] All of this can occur while a single machine axis is being actuated. However, it is also possible to actuate two or more machine axes in a coupled manner, thus capturing the machine's behavior when two or more axes are controlled simultaneously. This can reveal, for example, amplified vibrations that are greater than would be expected based solely on the vibration behavior when acting on a single machine axis, or it can detect controller errors that only become apparent when two machine axes are actuated synchronously. Status data

[0059] The monitoring device 44 determines various status data from the received measurement data. This status data allows direct or indirect conclusions to be drawn about the condition of the machine or its individual components. The status data includes, in particular, spectral data obtained from the measurement data through spectral analysis. Complete spectra or only the spectral intensities at selected discrete excitation frequencies can be determined.

[0060] The Fig. 2 This shows an example of a spectrum that can be obtained from a time signal of an acceleration, position, or current sensor, recorded during the actuation of a machine axis (here the B-axis, i.e., the tool spindle), through filtering and an FFT operation. The spectrum of Fig. 2 contains several clearly visible peaks at integer and non-integer multiples of the rotation frequency (orders) of the machine axis in question.

[0061] For example, strong peaks in tool speed and its multiples can indicate runout errors in the tool spindle. Peaks at higher multiples of the tool speed can indicate bearing damage in the tool spindle, and the bearing orders can potentially be deduced from the multiples. If the bearing orders are known, it may be possible to identify the bearing causing the peaks.

[0062] The monitoring device 44 transmits the status data thus obtained to the service server 45. EOL check

[0063] The finished workpieces are each installed in a gearbox. Before delivery, the gearbox is tested on an end-of-life (EOL) test bench. This is done using the... Fig. 3 explained in more detail.

[0064] The Fig. 3Figure 1 shows, in a highly schematic manner, machine 1 with the various servers 45-47 and the terminal device 48, which have already been described above. Also shown is a highly schematic representation of the EOL test bench 2. The EOL test bench communicates with the service server 45 via the web server 47.

[0065] As already explained, machine 1 has a variety of sensors, including acceleration sensors (vibration sensors) 51, position sensors 52 and current sensors 53. As also already explained, the machine uses these sensors to acquire measurement data and sends derived status data to the service server 45.

[0066] The EOL test bench also features a variety of sensors, including acceleration sensors 54, which measure acoustic signals while the installed workpiece rolls on a mating gear in the gearbox, rotary angle sensors, etc. The EOL test bench calculates EOL data from this through spectral analysis and also sends this data to the service server 45. Service server

[0067] Service server 45 processes the received data, calculates further values ​​if necessary, and stores the received data and any calculated values ​​in the database DB. Specifically, the service server stores the following data: Condition data with associated condition identification data that allows for the unique identification of the machine on which the condition data was determined and the associated operation in the test cycle (in particular the actuated machine axis), as well as the time of the test cycle; process data of the machining process for each workpiece, together with workpiece identification data that allows for the unique identification of the workpiece; EOL data from the EOL test bench, together with the associated workpiece identification data.

[0068] The service server can read and combine data from the database. In particular, the service server can combine EOL data for a specific workpiece with the associated process data and those machine status data that best characterize the machine state in relation to the machining state, each into a single data record. Prediction of disturbing noise excitations in the EOL spectrum based on kinematic links

[0069] The service server can perform a qualitative prediction of the intensities of disturbance orders on the EOL test bench. To do this, the service server calculates the intensity from the spectrum of Fig. 2 a corresponding expected excitation spectrum on the EOL test bench (EOL spectrum).

[0070] In its calculations, the service server utilizes the known kinematic links between the components of machine 1. This is demonstrated using the Figures 4A and 4B explained in more detail.

[0071] The Fig. 4AFigure 1 shows an excerpt from a table listing known possible disturbance orders of the B-axis (i.e., the tool spindle) and the corresponding expected disturbance orders in the end-of-line (EOL) spectrum. In this example, these disturbance orders are in a fixed ratio of 3.45, which is determined by the kinematic connection between the tool spindle and the workpiece, i.e., by the rolling coupling between the tool and the workpiece, and is further defined by the geometry of the workpiece and the grinding worm. Figuratively speaking, this ratio indicates how vibrations of the B-axis are transmitted as waviness on the tooth flanks of the workpiece. This ratio can be calculated by considering the engagement conditions between the grinding worm and the workpiece. It is referred to below as the "transmission factor." The possible disturbance orders of the B-axis can be determined if the orders of the B-axis components are known, e.g.,Bearing and motor orders are calculated in advance. Actual disturbance orders of the B-axis can be determined by measurements.

[0072] The Fig. 4BFigure 1 shows an excerpt from a table listing possible disturbance orders of the Y-axis (i.e., the shift axis) and the corresponding expected disturbance orders in the end-of-line (EOL) spectrum. The table distinguishes between different components of the Y-axis, such as the ball screw drive and the drive motor, which can cause these disturbance orders, and between disturbance orders in the EOL spectrum resulting from vibrations during workpiece machining (grinding) and dressing. Vibrations during grinding directly lead to flank waviness on the workpiece flanks. Vibrations during dressing initially manifest as flank waviness on the grinding worm and are then translated into flank waviness on the workpiece flanks during grinding.The corresponding transfer factors between possible disturbance orders of the Y-axis and the resulting disturbance orders in the EOL spectrum can also be easily calculated if the kinematic relationships and the machining parameters during grinding and dressing are known. The possible disturbance orders of the Y-axis can, in turn, either be measured or calculated.

[0073] This type of analysis of possible disturbance orders of a machine axis and the resulting disturbance orders in the EOL spectrum can be performed for every machine axis involved in the grinding process.

[0074] The prediction of an EOL spectrum is now based on the spectra determined during the test cycle on the machine and the known transfer factors between disturbance orders of the machine axes and associated disturbance orders in the EOL spectrum. This is done using the Fig. 5 explained in more detail. Fig. 5shows a predicted EOL spectrum that is expected when the B-axis test in the test cycle shows the spectrum of Fig. 2 The predicted EOL spectrum essentially corresponds to the spectrum of the Fig. 2 However, it is stretched along the horizontal axis by the transfer factor of 3.45 mentioned above as an example. The absolute signal values ​​in this EOL spectrum should be viewed with caution: the actual strength of an EOL signal for a given disturbance order depends not only on the strength of the disturbance order of the machine axis causing it, but also on a multitude of other factors during workpiece machining and the workpiece's installation conditions in the gearbox. Therefore, the spectrum of Fig. 5It does not provide quantitative information about expected signal strengths. However, it allows a prediction of which disturbance orders will be present in the EOL spectrum based on the disturbance orders already present in the spectrum of the machine axis in question, and it allows a qualitative estimation of the expected signal strengths for these disturbance orders. For example, the spectrum of Fig. 5 A rough estimate of the signal strengths for certain interference orders of interest that cause noises perceived as particularly unpleasant. Such interference orders are found in the Fig. 5 For example, marked with a circle.

[0075] Overall, this allows for a good prediction of which disturbance orders with approximately which signal strengths are to be expected in the EOL spectrum due to which causes.

[0076] For example, a worn bearing in the tool spindle can cause vibrations in the tool spindle, the order of which (relative to the tool rotation) is determined by the bearing order. The bearing order results from the bearing design and can often be obtained from the bearing manufacturer. Therefore, vibrations measured during a test cycle can sometimes be directly attributed to the worn bearing. These vibrations can be measured, for example, by an accelerometer on the workpiece spindle housing. The vibrations are transmitted to the workpieces during the machining process and manifest there as periodic deviations (ripples) on the tooth flank. After installation in a gearbox, these ripples become noise excitations when the workpiece teeth roll against a mating tooth.The order of these noise excitations, relative to the workpiece rotation in the gearbox, can be easily calculated based on the above considerations. In this way, it is possible to calculate how the worn bearing will affect the noise spectrum of a gearbox.

[0077] Although the calculated spectrum of Fig. 5 This alone does not provide any quantitative information about expected signal strengths. However, by observing how this expected spectrum changes from test cycle to test cycle, a very good estimate can be made of which disturbance orders change and in what way. This will be shown below using the Fig. 6 Explained using an example. Fig. 6The diagram shows the expected spectral intensity I in the EOL test bench at a specific order (here, order 52) as a function of the number of workpieces processed by the machine. It is evident that the expected noise intensity increases significantly over time. By fitting this to a suitable regression function (here, a quadratic regression function), this increase can be quantitatively measured, and depending on the determined regression parameters, an appropriate action can be triggered, such as issuing a warning signal. Furthermore, the temporal evolution of the expected signal strengths at the different EOL orders can be visualized in a suitable manner. This enables even inexperienced users to interpret noise problems. Identification of interference sources based on kinematic links

[0078] The reverse approach is also possible: If an EOL spectrum has been determined through measurements on the EOL test bench, the above considerations can be used to estimate which machine axes, and possibly even which components of a machine axis, caused the disturbance orders in the measured EOL spectrum. This involves calculating backwards to determine which machine axis orders correspond to the disturbance orders in the measured EOL spectrum, and then identifying the component whose expected disturbance orders in the spectrum of a machine axis correspond to the orders calculated in this way. This process can be easily automated. Proceeding without knowledge of the kinematic connections

[0079] If the kinematic links of the powertrain are unknown or should not be used for calculation for other reasons, it is possible to predict signal intensities at certain EOL disturbance orders or to identify disturbance sources using a machine learning algorithm (ML algorithm).

[0080] This will be demonstrated below using the Fig. 7 This is explained. This is a highly simplified schematic diagram of an artificial neural network (ANN). In this example, the network has only three inputs and two outputs, as well as a single hidden layer. In reality, an ANN will usually have more inputs, outputs, and hidden layers.

[0081] In this example, the ANN receives state data at its inputs, each characterizing the oscillation tendency of one of the machine's axes B, Y, and Z. From this, the ANN calculates predicted end-of-line (EOL) spectral data in the form of expected spectral intensities at two specific EOL orders, here orders 52 and 59.

[0082] The ANN was previously trained with training data. In the Fig. 8 An example of such training data is shown. Fig. 8The table shows, on the one hand, condition data obtained from numerous machine test cycles. On the other hand, it contains end-of-life (EOL) data in the form of spectral intensities at orders 52 and 59, obtained through EOL measurements on gearboxes. Workpieces were installed in these gearboxes and machined while the machine was in the state in which the condition data was determined (i.e., shortly before and / or after the respective test cycle). The table contains a large number of such rows. It can be retrieved from the database DB of the Figure 1 and 3 The ANN was trained on this data in a known manner. This enables it to reliably predict which machine states (represented by state data) will lead to which EOL intensities at the specified orders.

[0083] The reverse direction is also possible: The input variables of a corresponding ANN can be EOL data, and the output data can be predicted state data.

[0084] Of course, the above example is highly simplified, but it demonstrates the basic procedure. Instead of an ANN, other types of ML algorithms can also be used, especially other well-known classifiers. Output of results and visualization

[0085] The visualization of the results of these analyses can be performed platform-independently on any client computer via a web browser. Other evaluation measures can also be implemented in a similarly platform-independent manner. This facilitates remote analysis. In particular, the condition of any machine can be checked in detail from any mobile device via the cloud.

[0086] Additionally, it is conceivable to automatically send a corresponding message via SMS, push notification or email when conditions arise that require intervention.

Claims

1. A method of monitoring a condition of a gear cutting machine having a plurality of machine axes, characterized in that the method comprises the steps of: a) performing a test cycle, wherein in the test cycle at least a portion of the machine axes is systematically actuated and associated machine measurement data are obtained; b) performing a spectral analysis of the machine measurement data, wherein machine spectral data are calculated from the machine measurement data; c) determining predicted EOL spectral data based on the machine spectral data, wherein the predicted EOL spectral data indicate at which orders excitations are to be expected in an EOL spectrum when a workpiece machined with the gear cutting machine is installed in a gear train and rolls off on a mating gear in the gear train; and d) optionally, outputting the predicted EOL spectral data or at least one quantity derived therefrom.

2. The method according to claim 1, wherein determining the predicted EOL spectral data comprises applying a propagation factor to the machine spectral data, the propagation factor depending on a kinematic linkage between the machine axis for which the machine spectral data was determined and the workpiece.

3. The method according to claim 1 or 2 wherein the predicted EOL spectral data are determined individually per actuated machine axis.

4. The method according to any of the preceding claims, wherein steps a) to c) are repeated several times, wherein workpieces are machined with the gear cutting machine between the test cycles and the test cycles are performed in machining pauses in which the machining tool is not in a machining engagement with a workpiece, and wherein a development of the predicted EOL spectral data as a function of the test cycles performed, the workpieces machined or the time is visualized and / or analyzed, in particular by a regression analysis.

5. The method according to any of the preceding claims, wherein reference machine spectral data are available for a plurality of reference machines, the reference machine spectral data having been determined by a plurality of reference test cycles performed on the reference machines, wherein predicted reference EOL spectral data are determined from the reference machine spectral data, wherein the predicted EOL spectral data, which have been determined based on the machine spectral data of the monitored gear cutting machine, are compared to the predicted reference EOL spectral data or quantities derived therefrom, wherein the method optionally comprises a statistical analysis of the predicted reference EOL spectral data.

6. A method of monitoring a condition of a gear cutting machine having a plurality of machine axes, characterized in that the method comprises the steps of: a) performing an EOL test on a gear train comprising a workpiece machined by the gear cutting machine, wherein in the EOL test the workpiece in the gear train rolls off on a mating gear and associated EOL measurement data are determined; b) performing a spectral analysis of the EOL measurement data, wherein EOL spectral data from the EOL measurement data are calculated; c) determining predicted condition data based on the EOL spectral data, wherein the predicted condition data for at least one machine axis indicates which orders of that machine axis are consistent with the calculated EOL spectral data; and d) optionally, outputting the predicted condition data or at least one quantity derived therefrom.

7. The method according to claim 6, further comprising: e) performing a test cycle in which at least a portion of the machine axes are systematically actuated and associated machine measurement data are obtained; f) performing a spectral analysis of the machine measurement data, wherein machine spectral data are calculated from the machine measurement data; and g) determining predicted EOL spectral data based on the machine spectral data, wherein the predicted EOL spectral data indicate at which orders excitations are to be expected in an EOL spectrum when a workpiece machined by the gear cutting machine is installed in a gear train and rolls off on a mating gear in the gear train, wherein determining the predicted condition data comprises comparing the EOL spectral data calculated from the EOL measurement data to the predicted EOL spectral data.

8. A method for creating a training data set of a machine learning algorithm for use in a method of monitoring a condition of a gear cutting machine having a plurality of machine axes, characterized in that the method comprises: a) performing a test cycle in which at least a portion of the machine axes is systematically actuated and associated condition data are determined by measurements; b) machining at least one workpiece with the gear cutting machine while the gear cutting machine is in a condition that corresponds to the condition data; c) installing the machined workpiece in a gear train; d) performing an EOL test on the gear train, wherein in the EOL test the workpiece in the gear train rolls off on a mating gear and associated EOL data are determined; e) storing the condition data and the corresponding EOL data in the training data set; f) repeating steps a) to e) for a plurality of test cycles and machined workpieces, wherein the workpieces have the same nominal geometry and are machined under the same machining conditions, wherein the machine learning algorithm is preferably a classification algorithm, in particular an artificial neural network or a support vector machine, or a random forest.

9. A method of training a machine learning algorithm for use in a method of monitoring a condition of a gear cutting machine having a plurality of machine axes, characterized in that the machine learning algorithm is trained using the training data set according to claim 8.

10. A method of monitoring a condition of a gear cutting machine with a plurality of machine axes, characterized in that a machine learning algorithm trained with the training data set according to claim 8 is used in the method.

11. The method according to claim 10, wherein the machine learning algorithm has condition data of the gear cutting machine as input variables and predicted EOL data as output variables, the method comprising: a) performing a test cycle, wherein in the test cycle at least a portion of the machine axes is systematically actuated and associated condition data are determined by measurements; b) determining predicted EOL data based on the condition data by feeding the condition data to the trained ML algorithm as input variables; and c) optionally, outputting the predicted EOL spectral data or at least one quantity derived therefrom.

12. The method according to claim 10, wherein the machine learning algorithm has EOL data as input variables and predicted condition data of the gear cutting machine as output variables, the method comprising: a) performing an EOL test on the gear train, wherein in the EOL test the workpiece in the gear train rolls off on a mating gear and associated EOL data are determined; b) determining predicted condition data based on the EOL data by feeding the EOL data to the trained ML algorithm as input variables; and c) optionally, outputting the predicted condition data or at least one quantity derived therefrom.

13. The method according to any one of claims 8-12, wherein the condition data correlate with a condition of a machine axis with respect to its vibration behavior and, in particular, comprise machine spectral data calculated by a spectral analysis of machine measurement data, and / or wherein the EOL data correlate with the noise behavior of the gear train and, in particular, comprise EOL spectral data calculated by spectral analysis of EOL measurement data.

14. A device for monitoring a condition of a gear cutting machine having a plurality of machine axes, comprising a processor (451) and a storage medium (452) on which is stored a computer program which, when executed on the processor, causes the following steps to be performed: receiving condition data determined by a test cycle of the gear cutting machine, wherein in the test cycle at least a portion of the machine axes has been systematically actuated and the associated condition data have been determined by measurements; and determining predicted EOL data correlated with a noise behavior of a gear train comprising a workpiece machined with the gear cutting machine, based on the condition data.

15. A device for monitoring a condition of a gear cutting machine having a plurality of machine axes, comprising a processor (451) and a storage medium (452) on which is stored a computer program characterized in that the computer program, when executed on the processor, causes the following steps to be performed: receiving EOL data determined by an EOL test on a gear train comprising a workpiece machined by the gear cutting machine, wherein in the EOL test the workpiece in the gear train rolls off on a mating gear and the associated EOL data was determined; and determining predicted condition data that correlates with a condition of at least one machine axis in terms of its vibration behavior, based on the EOL data.