METHOD AND SYSTEM FOR DETERMINING THE DYNAMIC BEHAVIOR OF A MACHINE

Description

Background of the invention

[0001] The invention relates to a method for determining the dynamic behavior of a multi-axis machine. The invention also relates to a corresponding system.

[0002] Measurements used to assess the dynamic properties of machines, especially machine tools, particularly complex frequency response measurements, are currently performed individually for each axis, depending on its position or pose. Therefore, it is only possible to make very limited statements about the dependencies of an axis's dynamic properties on its own axis position and on the positions of other axes, as well as to identify complex interrelationships.

[0003] In particular, current state-of-the-art methods only perform measurements of dynamic properties at a few selected positions within the workspace. This severely limits the validity of statements about spatial dependencies. If spatial dependencies with respect to other axes are to be investigated, this is only possible by preconfiguring a few starting positions. Even then, only spatially discrete statements are possible.

[0004] From EP 3 425 460 A1 a method for determining the condition of a spindle of a machine tool is known.

[0005] A method for diagnosing a machine spindle is known from US patent 2019 / 0310605 A1.

[0006] From EP 3 396 485 A1 a method for evaluating at least one mechatronic characteristic of a machine tool is known.

[0007] From EP 3 168 700 A1 a method for the automated monitoring of a CNC-controlled multi-axis machine is known. Object of the invention

[0008] The object of the present invention is to determine the dynamic behavior of a machine with multiple axes over the entire working space. Description of the invention

[0009] This problem is solved according to the invention by a method according to claim 1 and a system according to claim 14.

[0010] In machine tools, machining units, such as tool holders that hold a tool or laser processing heads, are moved in a specific axial direction via drives and potentially interposed mechanical components, such as gearboxes or gantries to which the machining unit may be attached. This is often referred to simply as an "axis." An axis position is thus the position of the machining unit in the axial direction caused by the corresponding axis. The working range of an axis is the area within which the machining unit can be moved by the respective axis. The workspace is created by the superposition of the working ranges of the axes of a machine tool. The workspace is therefore the area of ​​space that can be reached by the machining unit. A pose or spatial position is the combination of the position and orientation of an object, e.g., a machining unit.

[0011] The inventive method performs measurement runs in such a way that the entire working area is covered. Measurement runs are carried out for each axis across the entire working range of the axis under investigation. In particular, multiple measurement runs can be performed for each axis.

[0012] During a test run, data from various sources is collected and recorded. For example, a machine may be equipped with sensors that perform measurements. The machine's controls and drives can also be data sources, and their input and / or output signals can be recorded.

[0013] The data acquired and recorded during a measurement run can either be recorded synchronously in time or synchronized later. The recording duration can depend on the axis length of the axis under consideration. Preferably, a short sampling time is selected for data acquisition, taking into account the system behavior to be observed (sampling theorem). The method according to the invention enables a comprehensive assessment of the spatial dependencies of a machine's dynamic properties with time (spatial) frequency resolution, limited only by the feed rate used for measurement and the sampling frequency of the data acquisition. In particular, statements can be made about the dynamic properties of an axis as a function of its own axis position and the position of other axes. Thus, global statements about the complete spatial dependencies are possible. Interrelationships can be automatically detected.

[0014] A measurement run for an axis can be performed in sections. In particular, different sections can be recorded at different times. However, the overall goal is to record the entire axis, i.e., the entire working range of the axis.

[0015] The measurement run can be performed during idle times or while the machine is running. Idle times are periods when the machine is not machining the workpiece. However, it is also conceivable that at least sections of an axis's working range are recorded concurrently with the main machining operation, for example, when positioning between contours, i.e., when the machining unit is moved from one machining point to another. Data recording during the machine's operation as online measurements is also possible.

[0016] Alternatively, the measurement runs can be carried out in a separate measurement program. This measurement program can be performed, for example, before the actual commissioning of a machine or during the machine's idle times.

[0017] The system behavior, particularly that of the data processing unit, can be determined from the time-frequency plot of the recorded data. Time-indexed measurements allow for the determination of spectra of the excitation and output signals. The ratio of these spectra provides a representation of the system's dynamic behavior in the frequency domain.

[0018] The analysis of the time-frequency representation can be automated.

[0019] The measurement run can be performed with a predefined input signal, in particular a drive signal, where the input signal comprises a predefined signal to which an excitation signal is superimposed. For example, a constant speed can be specified as the predefined signal, to which an excitation signal is superimposed. An excitation signal can, for example, be a technical approximation of white noise. Depending on the type of measurement run or measurement to be performed, a specific, suitable speed profile can be specified.In the types of measurements used to assess dynamic properties (diagnosis of mechanical properties or the interaction of control engineering settings and mechanical properties) (for example, reference frequency response of position control loop, reference frequency response of speed / velocity control loop, disturbance frequency response of speed / velocity control loop, speed / velocity controller system, mechanical frequency response), a constant velocity can be selected as the input signal, to which a suitable excitation signal is superimposed.

[0020] To diagnose the components involved in the movement of each axle's drive train (e.g., drive, rack and pinion, measuring systems), the superimposed excitation signal can be set to zero. Several individual measurement runs can be performed with different feed rates, i.e., speeds, or with other suitable speed profiles (e.g., a linear increase in speed across the axle position).

[0021] To investigate dependencies on the workspace position, a multi-stage approach is conceivable. To account for the dependencies of the dynamic properties on the workspace position of an axis other than the one under investigation, these can be pre-positioned at specific positions, and the measurement / test run can be performed for each pose. Thus, an axis other than the one under investigation can be pre-positioned in different positions, and a test run can be performed for the axis under investigation for each position of the other axis.

[0022] A measurement run can be performed for at least one axis while another axis is moving. In particular, the simultaneous movement of multiple axes in conjunction with the recording of data related to the axis under investigation is possible. Specifically, with several axes moving simultaneously, their mutual influence can be investigated by superimposing the excitation signals and recording the input and output signals, thus creating a representation of all moving (mechanical) components of the overall (machine) system. Several forms are conceivable for this representation. Characterization can be performed in the frequency domain as a frequency response. Furthermore, time-domain system identification methods, such as NARMAX (Nonlinear AutoRegressive Moving Average with eXogenous input), Subspace System Identification, etc., can be used.In particular, a coupling matrix can be created for linear time-invariant systems.

[0023] According to the invention, when measuring a single axis, the input and output signals are recorded as data across the entire working range of the axis. Output signals can include, for example, current (especially current drawn by the axis drive), feed rate, rotational speed (of the drive or, if a gearbox is present, of the output), acceleration, etc. To investigate the coupling of multiple axes, the number of data points to be recorded increases according to the transmission behavior under investigation.

[0024] Data acquisition can be achieved, for example, via a data acquisition unit capable of recording data synchronously throughout the entire measurement period, particularly during a test run. The data acquisition unit can be part of a control system (of the machine) and interact with a storage unit. Alternatively, the data acquisition unit can be an external component, such as an industrial PC (IPC), or a real-time capable data acquisition unit with at least one physical communication interface and, in particular, with data storage functionality connected to the data sources, enabling data recording and storage.

[0025] The time-frequency representation can be determined, for example, using a transformation, in particular a Fourier transform, quadratic transform, or Wigner-Ville distribution. A short-time Fourier transform, for instance, can be performed. For at least one axis, the time reference can be converted into a spatial reference. The analysis and evaluation of the representation in the image domain is performed using image processing algorithms based on defined indicators, such as statistical values ​​for individual features like thresholds with / without spatial reference. Furthermore, the analysis can be performed using the representation itself as a highly correlated tensor, which is then further processed in the data processing unit. For example, the data from the Bode plot can be stored in two matrices, where the amplitude information is stored separately in one matrix and the phase information in another, each converted into a normalized gray value depending on the location.This makes it possible to analyze the data using image processing algorithms and to train neural networks in order to replace the previous series-dependent, locally valid examination methods with holistic procedures. This leads to a significantly simplified diagnostic process.

[0026] Further parameters such as component lifespan and wear, as well as service calls, can be considered in the data analysis. This analysis can be performed with respect to a population or with respect to the temporal progression of the same or similar machines per characteristic, which also enables the automated identification of relevant characteristics. The population can be the set of systems (i.e., machine tools) with nominally identical characteristics (i.e., motor, gearbox, machine body, etc.). The "characteristics" extracted from measurements can be, for example, natural frequencies, damping, hysteresis, etc. These characteristics are statistically distributed. The characteristics of these distributions (e.g., means, standard deviations) can, in turn, be relevant for characterizing a population.

[0027] For example, if the standard deviation of a natural frequency is greater for machines from production line A than for those from production line B, this indicates differences in the (nominally identical) production process.

[0028] The image processing algorithms can incorporate machine learning. In particular, the algorithms can build a statistical model based on training data, i.e., the acquired data. Specifically, patterns and regularities in the acquired data can be recognized. Thus, even unknown data can be evaluated after the learning process is complete. One possible applied learning variant using artificial neural networks is so-called deep learning, which can be applied according to the invention. Examples of deep learning include convolutional neural networks, deep auto encoders, and generative adversarial networks.

[0029] In particular, it may be provided that an artificial neural network is trained, in which the analysis of the representation in the image area is fed to the neural network.

[0030] Metadata can be provided to the neural network. In particular, information from other data sources, such as data labeling based on service history (e.g., information about tasks performed or components replaced), can be fed into the neural network. This metadata can be considered during analysis for the automatic classification of measurement data. This enables the automated recognition of complex relationships.

[0031] At least some of the process steps can be performed on spatially distributed systems. For example, the image processing algorithm can be executed on the machine, on a real-time capable data acquisition unit with at least one physical communication interface and, in particular, with data storage functionality, on a computer, or in a cloud infrastructure.

[0032] The invention also encompasses a system for determining the dynamic behavior of a multi-axis machine, comprising a machine, in particular a machine tool, having multiple axes, at least one data source connected to a data acquisition unit, a data storage unit for storing the acquired data, a data processing unit configured to determine a time-frequency representation of the recorded data, and an image processing unit for processing the time-frequency representation. The data sources can be internal or external to the machine. Internal data sources can, for example, provide current, acceleration, velocity, actual position, and / or setpoint values. Furthermore, internal data sources can originate from controllers, such as a programmable logic controller (PLC) or a numerical control (NC).Furthermore, the drives of the axes can represent data sources. Additionally, data sources such as laser power, gas pressure, scattered light, etc., can be provided as data.

[0033] External data sources can include microphones, micro-electro-mechanical systems (MEMS) sensors, or cameras. The data acquisition unit can be part of a controller. Alternatively, the data acquisition unit can be an external component for data acquisition, such as a real-time data acquisition unit with at least one physical communication interface and, in particular, with data storage functionality, or an industrial PC (IPC).

[0034] The image processing system can be configured as a neural network or incorporate one. With such a system, it is possible to employ self-learning image processing algorithms, particularly machine learning and deep learning. Furthermore, it is possible to classify the data based on additional information, such as service reports, spare parts replacements, etc. This allows for a comprehensive assessment of the spatial dependencies of a machine tool's dynamic properties with time-(spatial)-frequency resolution, limited only by the feed rate used for measurement and the sampling rate during data acquisition. Local deviations of the system under investigation can be detected. Moreover, a global assessment of spatial dependencies can be made in a significantly shorter time than previously possible. Complex interrelationships can be automatically identified.

[0035] Further features and advantages of the invention will become apparent from the following detailed description of an embodiment of the invention, with reference to the figures in the drawing, which show essential details of the invention, and from the claims. The features shown therein are not necessarily to scale and are depicted in such a way that the inventive features are clearly visible. The various features can be implemented individually or in any combination in variants of the invention.

[0036] The schematic drawing shows an embodiment of the invention, which is explained in the following description.

[0037] They show: Fig. 1 a system according to the invention for determining the dynamic behavior of a machine; Fig. 2 a spatial-frequency representation of data; Fig. 3 a flowchart of the method according to the invention.

[0038] Fig. 1Figure 10 shows a system 10 for determining the dynamic behavior of a machine 11, which has several axes 12, 13. The machine 11 also has data sources 14, 15, 16, where data sources 14, 15 are assigned to axes 12, 13. These can be internal sensors. Data sources 14, 15 can also be controllers. The data output can include, for example, current or rotational speed. The machine 11 can have additional data sources 16, which are, for example, designed as sensors. For example, the acceleration of a machining unit can be detected in this way. Additionally, external data sources 17 can be provided. Data source 17 can be, for example, a microphone or camera. By using additional data sources, such as accelerometers, gyroscopes, microphones, cameras, etc., characteristic system properties can be determined.

[0039] The data acquired from data sources 14, 15, 16, and 17 are fed to a data acquisition unit 18. The data from data sources 14, 15, 16, and 17 can be recorded in a time-synchronized manner; in particular, data sources 14-17 can be synchronized. Alternatively, the data can be synchronized subsequently, for example, in the data acquisition unit 18. Specifically, the data acquisition unit 18 can be configured to record the data in a time-synchronized manner over the entire measurement period. The data acquisition unit 18 can be part of a control system. In the illustrated embodiment, it is designed as an external, real-time capable data acquisition unit with at least one physical communication interface and, in particular, with data storage functionality.

[0040] During data acquisition, an axis 12, 13, or machine 11 is moved through its entire working range. A possible input signal is a constant speed for a single axis of the system (a predefined constant speed for the movement of a machining unit along the axis), superimposed with an excitation signal. The output signal, i.e., the data supplied by data sources 14-17, can correspond, for example, to the current, feed rate, rotational speed, or acceleration.

[0041] To investigate the coupling of multiple axes 12, 13, the number of data points to be recorded increases. The recorded data are stored in a data storage unit 19. Subsequently, data processing or preprocessing takes place in a data processing unit 20. Here, a time-frequency representation of the recorded data is determined. Alternatively or additionally, the system behavior of the machine 11 can be determined, for example, as a transfer function assuming a linear time-invariant system or as a representation using the NARMAX approach. Optionally, the time-frequency representation can then be converted into a spatial-frequency representation.

[0042] The data is then analyzed in an image processing unit 21. In particular, the representation in the image area resulting from the time-frequency transformation can be evaluated using defined indicators. For example, statistical values ​​for individual features, such as thresholds with and without spatial reference, can be determined. The representation itself can be further processed as a highly correlated tensor. One approach, for example, is to represent the information from the Bode plot in two matrices, with the amplitude information being converted separately in one matrix and the phase information in another into a normalized gray value depending on location.This makes it possible to analyze the data using modern image processing algorithms, particularly machine learning, and to train suitable neural networks to replace previous series-dependent, locally valid examination methods with holistic procedures. This leads to a significantly simplified diagnostic process, including the determination of the properties relevant for the examination.

[0043] The image processing unit 21 can be supplied with so-called metadata via an additional memory 22. This allows further parameters to be taken into account during data analysis. These additional parameters include, for example, the lifespan of a machine, wear and tear, service interventions, etc.

[0044] The Fig. 2This shows a spatial frequency diagram. This diagram was determined by measuring the frequency response of the x-axis of a machine while the y-axis moved. The colors or shades of gray represent the amplitude of the frequency response. This representation allows the frequency response of the x-axis to be determined at any time and at any point. Thus, instead of only having point-source data, a global frequency response can be determined across the entire working range of the x-axis.

[0045] The Fig. 3 Figure 1 shows a flowchart of the method according to the invention. In step 100, at least one measurement run is carried out for each axis over its entire working range.

[0046] In step 101, data related to the measurement run are acquired and recorded. In step 102, a time-frequency representation of the recorded data is determined using a data processing unit. In step 103, the time-frequency representation, or a related representation such as a position-frequency representation, is analyzed using an image processing algorithm.

Claims

1. A method for determining the dynamic response of a machine (11) having at least one axis (12, 13), comprising the method steps: a. carrying out a measurement run for each axis (12, 13) over the entire working range thereof, wherein each axis is moved over its entire working range during data collection; b. collecting and recording data that relate to the measurement run, wherein data from different data sources (14, 15, 16, 17) are collected and recorded during the measurement run; c. determining, by means of a data processing unit (20), a time-frequency representation of recorded data; d. evaluating the time-frequency representation, or a location-frequency representation related thereto, by means of an image processing algorithm with the aid of established indicators.

2. The method according to claim 1, characterized in that the measurement run for an axis (12, 13) takes place in sections.

3. The method according to any of the preceding claims, characterized in that the measurement run is carried out during downtimes or while the machine (11) is operating.

4. The method according to any of the preceding claims, characterized in that the measurement run is carried out with a predefined input signal, wherein the input signal comprises a predefined signal with an excitation signal superimposed thereon.

5. The method according to any of the preceding claims, characterized in that for at least one axis (12, 13) a plurality of measurement runs are carried out with different input signals.

6. The method according to any of the preceding claims, characterized in that for at least one axis (12, 13) a measurement run is carried out while another axis is being moved.

7. The method according to any of the preceding claims, characterized in that another axis (12, 13) is pre-positioned in different positions, and a measurement run is carried out for each position of said other axis (12, 13).

8. The method according to any of the preceding claims, characterized in that the time-frequency representation is determined by means of a transformation, in particular a Fourier transformation, quadratic transformation or Wigner-Ville distribution.

9. The method according to any of the preceding claims, characterized in that for at least one axis (12, 13), the time reference is converted to a location reference.

10. The method according to any of the preceding claims, characterized in that the evaluation of the representation in the image region takes place by means of image processing algorithms.

11. The method according to any of the preceding claims, characterized in that an artificial neural network is trained by feeding the analysis of the representation in the image region to the neural network.

12. The method according to any of the preceding claims, characterized in that metadata are made available to the neural network.

13. The method according to any of the preceding claims, characterized in that at least some of the method steps are carried out on spatially distributed systems.

14. A system (10) for determining a dynamic response of a machine (11) having a plurality of axes (12, 13), the system comprising a machine (11) with a plurality of axes (12, 13), at least one data source (14 - 17) connected to a data collection unit (18), a data storage unit (19) for storing the collected data and a data processing unit (20) that is configured to determine a time-frequency representation of recorded data, as well as an image processing device (21) for processing the time-frequency representation, wherein the system (10) is configured to carry out a method according to any of the claims 1 to 13.

15. The system according to claim 14, characterized in that the image processing device (21) is designed as a neural network or comprises the same.