Method and device for monitoring a milling machine

Abstract

A method for monitoring a milling machine (1), comprising the following steps: - deploying (S1) an untrained machine learning model (M) for determining one or more anomalies in time series data; - obtaining (S2) first time series data (I, II) representing the rotation speed of a milling head of the milling machine (1) and at least one other operating parameter of the milling machine (1) by the untrained machine learning model (M), preferably from a converter that powers the milling machine (1) during operation of the milling machine (1); and - during operation of the milling machine (1), training (S3) the untrained machine learning model (M) based on the obtained first time series data (I, II); - during operation of the milling machine, obtaining (S4) second time series data (III) representing the rotation speed of the milling head and other operating parameters of the milling machine (1) by the trained machine learning model (M); - during operation of the milling machine (1), determining (S5) one or more anomalies in the second time series data (III) by the trained machine learning model (M).

Description

Technical Field

[0001] The invention relates to a method and a device for monitoring a milling machine, in particular by means of machine learning. Background Art

[0002] Patent application US 2016 / 091393 A1 describes a method or system for forecasting and planning a production facility. To detect tool wear, the performance of the spindle is monitored, among other things. To compensate for tool wear, the performance of the spindle and therefore its rotation speed is increased. The long-term trend of the spindle performance is extrapolated in order to predict the remaining service life.

[0003] Patent application US 2007 / 088550 A1 describes a method for predictive maintenance of a machine. To this end, data related to the operation of the machine, such as vibrations, speed or current, are recorded. So-called features are generated from these data in order to make error predictions.

[0004] Patent application US 2004 / 179915 A1 describes a machine tool. In order to analyze the behavior of the machine tool, signals known as characteristic variables are recorded there and used as input variables for an artificial neural network.

[0005] US 8781982 B1 describes a method and a device for determining the remaining service life. For this purpose, operating data are fed into an artificial neural network. The operating data may be vibration data, acoustic data and acceleration data.

[0006] In addition, European patent application EP19166079 and international patent application PCT / EP2020 / 058069 disclose methods and devices for monitoring milling processes through artificial intelligence. Summary of the invention

[0007] Milling is generally a cutting manufacturing process for producing workpieces with geometrically defined shapes. During the milling process, material is removed by rotating the milling tool or milling head at high speed around its own axis, while the milling tool itself follows the contour to be produced on the workpiece or the workpiece moves accordingly relative to the milling head. Milling machines can be used to mill different workpieces, especially also for milling printed circuit boards. For example, a dividing line can be produced between the conductor track surfaces of a circuit board by milling. A large amount of dust is generated during the milling process, so that the milling spindle will be stuck after a certain running time. In traditional milling machines, such errors or faults are only detected after the corresponding errors (such as the jamming of the milling spindle) occur. The failure of the milling machine during the milling process can cause a significant delay in the manufacturing process of the object to be manufactured. In addition, the workpiece that is not completed in this manufacturing step is in many cases a defective product and cannot be further processed or may need to be tested for its normal function. In the ongoing manufacturing process, the stoppage of the milling machine requires the immediate use of maintenance personnel or service technicians to perform maintenance or repair measures. In many cases, such service technicians are not immediately available and must be requested. This may cause the entire production line in the factory to temporarily stop.

[0008] Furthermore, the milling machines can be operated in different modes, for example depending on their specific application. For example, in a production line, a plurality of milling machines can be operated, wherein the operation of each milling machine is affected by individual conditions, for example depending on the wear of the individual milling heads.

[0009] It is therefore an object of the present invention to provide a method and a device for monitoring a milling process which predicts possible faults due to the affected milling machine at an early stage, so that countermeasures can be taken in good time. That is, before the milling machine is damaged. By identifying faults at an early stage, the time of stopping the milling machine can be minimized. Another object of the present invention is to be able to monitor the milling machine independently of its type and independently of the specific application (e.g. workpiece production) for which it is used.

[0010] This object is achieved by the following aspects.

[0011] According to a first aspect, the object is achieved by a method for monitoring a milling machine. The method comprises the step of deploying an untrained machine learning model for determining one or more anomalies in time series data. The method further comprises obtaining, by the untrained machine learning model, during the operation of the milling machine, first time series data representing the rotation speed of a milling head of the milling machine and at least one other operating parameter of the milling machine, preferably from a converter that powers the milling machine, the at least one other operating parameter preferably corresponding to the power supply current or depending on the power supply current of the milling machine. The method further comprises training the machine learning model based on the obtained first time series data during the operation of the milling machine. The method further comprises obtaining, by the trained machine learning model, during the operation of the milling machine, second time series data representing the rotation speed of a milling head of the milling machine and another operating parameter. The method further comprises determining, by the trained machine learning model, one or more anomalies in the second time series data during the operation of the milling machine.

[0012] According to a second aspect, the object is achieved by a computer program product. The computer program product comprises a program code for performing the steps of the first aspect when executed.

[0013] According to a third aspect, the object is achieved by a device (eg, a milling machine, an edge device and / or a cloud platform) preferably comprising a processor and a memory. The device is configured to perform the steps of the first aspect. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 A flow chart is shown for illustrating a possible embodiment of monitoring a milling process.

[0015] Figure 2 A block diagram illustrating an exemplary embodiment of a milling machine is shown.

[0016] Figure 3 A block diagram illustrating another possible embodiment of a milling machine is shown.

[0017] Figure 4 A plurality of milling machines are shown, each of which may be monitored, according to one embodiment.

[0018] Figure 5 The operating modes of different milling axes are shown, wherein the rotational speed (x-axis) and the current are indicated.

[0019] Figure 6 The operating modes of the various milling axes are shown after the acceleration and deceleration phases have been removed.

[0020] Figure 7 First and second time series data of a milling machine are shown.

[0021] Figure 8 Exemplary method steps of an embodiment are shown.

[0022] Fig. 9 Exemplary method steps of another embodiment are shown.

[0023] Fig.10 Exemplary method steps of another embodiment are shown.

[0024] Fig.11 Exemplary method steps of another embodiment are shown.

[0025] Fig.12 Exemplary method steps of another embodiment are shown.

[0026] Fig.13 Exemplary method steps of another embodiment are shown.

[0027] Fig.14 Exemplary method steps of another embodiment are shown.

[0028] Fig.15 Exemplary method steps of another embodiment are shown.

[0029] Fig.16 Exemplary method steps of another embodiment are shown. DETAILED DESCRIPTION

[0030] From the basis Figure 1 As can be seen from the flowchart, in the illustrated embodiment, the method for monitoring a milling machine may include a plurality of steps.

[0031] In a first step S01, the rotational speed or the rotational speed of the milling head 2 of the milling machine 1 is recorded. In addition, in step S01, at least one further operating parameter of the milling machine 1 is recorded during the milling process. The further operating parameter has, for example, a power supply current for operating the milling machine 1. The power supply current is used, for example, to supply power to the electric motor 3 driving the milling head 2 of the milling machine 1.

[0032] In a further step S02 , the detected rotational speed R and the detected operating parameters, in particular the supply current I, are evaluated by a trained machine learning model for identifying anomalies A in the milling process.

[0033] In another possible step ( Figure 1(not shown), after the abnormality is detected, a message is generated for the operator of the milling machine 1. For example, the identified abnormality can be displayed in real time on the display unit of the milling machine 1. In addition, in another step, maintenance measures for eliminating the possible causes of the identified abnormality A can be automatically started. After the abnormality is detected in the milling process in step S2, the failure probability or possible failure cycle of the relevant milling machine 1 is calculated in one possible embodiment. Possible maintenance measures can be customized according to the calculated failure probability and / or expected failure cycle. For example, a milling machine 1 with a high probability of failure is serviced or repaired with high urgency or priority. In addition, based on the calculated expected downtime cycle, it can be controlled which maintenance measure is performed on which milling machine 1 in which cycle. In another possible embodiment, after the abnormality has been identified in the milling process, at least one possible cause of the occurrence of abnormality A is determined, such as wear or blockage of the drive spindle.

[0034] In one possible embodiment, the machine learning model used to determine anomalies may include a one-step-ahead predictor such as ARIMA, a spatial detector such as SVM, or Threshold value. Of course, other machine learning models such as artificial neural networks can be used. Preferably, the machine learning model is trained in a training phase after the start of the milling process. This makes it possible to run production without interruption and to simultaneously update the monitoring of the milling machine or the milling process. For this purpose, training data about the rotation speed R and at least one further operating parameter P of the milling machine 1 are obtained during operation of the milling machine.

[0035] In one possible embodiment, the machine learning model for identifying anomalies can be downloaded as an application (app) from a database via a network to a computing unit of the milling machine 1 and trained during the milling process, i.e., in a training phase during the operation of the milling machine. In one possible embodiment, the workpiece W to be milled by the milling machine 1 has a printed circuit board.

[0036] In a possible embodiment, in step S01, the detection of the rotation speed R and further operating parameters, in particular the supply current I, is performed based on corresponding data, which are fed to the machine learning model at a data rate of about 1 / sec (i.e. 1 Hz). The determination or identification of anomalies by the trained machine learning model in step S02 preferably takes place in real time during the milling process, i.e. during the production process.

[0037] In particular, the detection of one or more (short-term) current peaks in the supply current, for example from about 5 to 10 seconds, is considered an anomaly. The supply current increases suddenly, followed by a slowly decreasing falling edge. Such an anomaly is caused, for example, by milling chips entering the milling spindle. There, the milling chips cause increased friction. By re-adjusting the supply voltage, the supply current increases (sudden increase) in order to maintain a constant speed or rotational speed of the milling head. If the milling chips are crushed by the rotational movement of the milling spindle, the supply current is normalized again (falling edge). This is done in order to keep the rotational speed of the milling head or milling spindle constant. In order to detect anomalies, the sampling rate of the supply current can also be appropriately selected so that current peaks or anomalies are not missed by undersampling the supply current. In addition, the sampling rate of the supply current can be selected to be low enough at the same time so that an excessive amount of data does not occur.

[0038] Finally, a quality indicator can be formed which indicates whether the detected measurement point of the supply current is no longer within the normal range for the milling machine. For example, the higher the value, the more likely the measurement point is outside the normal range. The quality indicator can be formed as a function of the rotation speed and the supply current. Based on the number of anomalies that have occurred, the remaining service life or failure of the milling machine can be determined or predicted.

[0039] According to one embodiment, in particular in order to determine or predict a failure or a remaining useful life of a milling machine, a milling process, for example by milling a printed circuit board, which generates milling chips, is monitored.

[0040] In another embodiment, the supply current is monitored to determine whether there is a brief or momentary deviation from a base value of the supply current and whether the supply current returns to the base value. Such monitoring can be done by a trained adaptive algorithm. In addition, or alternatively, a threshold value can be provided, by which such a current peak of the supply current is identified. Once such a current peak in the supply current is detected, it can be monitored whether the supply current drops below the threshold again, in particular to the base value of the supply current. The base value can be the value of the supply current at which the current peak occurs. The base value of the supply current can be determined by the required rotational speed of the milling machine, more precisely by the spindle of the milling machine, or the rotational speed of the milling machine is determined by the base value of the supply current. In addition, the frequency of such current peaks can be recorded and, depending on this, a failure or a remaining service life of the milling machine can be determined; this can be done, for example, by a quality indicator.

[0041] The supply current needs to be sampled sufficiently to ensure that current peaks are identified. For example, the supply current may be sampled in such a way that current peaks caused by milling chips that have entered the milling spindle can be detected. For example, the supply current may be sampled approximately once per second or more frequently than once per second.

[0042] In another embodiment, it is monitored whether there are changes in the supply current at a (substantially) constant rotational speed, in particular current peaks. Such changes in the supply current are caused, for example, by the grinding of milling chips. Therefore, in particular at a constant rotational speed of the milling head or the milling spindle, the current peaks of the supply current generated by the grinding of the milling chips can be monitored. These current peaks can have a typical profile. For example, the typical characteristics of the above-mentioned current peaks can be monitored and / or detected by a trained machine learning model. For this purpose, the machine learning model is trained during the operation of the milling machine, since each milling machine can be operated in different operating modes, can be parameterized to perform specific tasks, etc. The current peaks can then be identified as abnormal by the trained machine learning model. In particular, the trained machine learning model can distinguish the current peaks caused by the grinding of milling chips from current peaks from other sources.

[0043] Furthermore, the machined workpiece can be a workpiece made of a homogeneous material or a portion of a workpiece machined by a milling machine, which portion consists of a homogeneous material. This ensures that a constant rotational speed of the milling head can be assumed. Otherwise, in the case of inhomogeneities in the workpiece, there would be variations in the rotational speed or the rotational speed would have to be adjusted depending on the cross section of the machined workpiece. It is thus possible to provide several portions that are machined at different predetermined target rotational speeds, i.e. speed set points. The respective target rotational speed can be used as a base value for detecting the current peak value.

[0044] Instead of or in addition to the supply current, the supply voltage can also be recorded and evaluated. Conversely, other variables derived from the supply current and / or supply voltage can also be used to monitor the milling process or to predict the remaining service life or probability of failure of the milling machine.

[0045] Printed circuit boards consist of an electrically insulating material with conductive connections (conductor traces) adhered to it. Fiber-reinforced plastic or, for cheaper devices, stiff paper are common insulating materials. The conductor traces consist primarily of a thin layer of copper.

[0046] Printed circuit boards can also be constructed by milling copper layers ("isolation milling", see the picture below for a soldering grid board). Such circuit boards do not include conductor tracks, but rather surfaces that are separated from one another by milling tracks.

[0047] In milling technology, a milling machine such as a milling cutter creates the dividing lines between the conductor surfaces. All copper remains in place (island excavation method). Wet chemistry and photolithography steps are omitted. CNC programs can be generated with specialized CAD software, allowing the circuit board to be manufactured quickly.

[0048] Figure 2 A block diagram illustrating an embodiment of a milling machine 1 according to the invention is shown. Figure 2The milling machine 1 shown is used for milling a workpiece W. The workpiece W is, for example, a printed circuit board. The milling machine 1 has at least one rotatable milling head 2 for machining the workpiece W during the milling process. The milling head 2 is driven by an electric motor 3 of the milling machine 1. Here, the electric motor 3 draws a supply current. The supply current may include one or more current phases (L1, L2, L3). Figure 2 As shown, the milling machine 1 according to the invention is operatively coupled to a machine learning module 4. The machine learning module 4 detects anomalies occurring during the milling process based on the detected rotation speed R of the milling head 2 and at least one other operating parameter I of the milling machine 1 during the milling process. The machine learning module 4 can evaluate several operating parameters I for detecting anomalies A. In one possible embodiment, the machine learning module 4 executes a trained machine learning model during the milling process.

[0049] The machine learning model of the machine learning module 4 is trained during the ongoing milling process. These training data may include, for example, power supply data and data on the rotation speed of the milling head 2. These training data may be located in a data memory. In a possible embodiment, the training data is loaded from a database of the milling machine operator and used for training purposes.

[0050] It is no longer necessary to select a machine learning model based on the type of milling machine 1 and / or the type of workpiece W to be milled during the milling process. In one possible embodiment, the machine learning model can also be downloaded from a database to the machine learning module 4 of the milling machine 1. For example, the machine learning model can be downloaded from the cloud platform 6 of the milling machine manufacturer or the milling machine operator via a data network. Alternatively, the machine learning model can be deployed in the cloud platform.

[0051] In one embodiment, the milling process itself can be controlled by a CNC program executed by a controller or microprocessor of the milling machine 1 according to the present invention. During the milling process, the feed movement of the milling head 2 is carried out perpendicularly or obliquely to the axis of rotation of the milling head 2. If the workpiece W is a printed circuit board, the milling machine 1 is controlled by software or a CNC program to generate a path around the desired conductor path. For example, the conductor trace to be manufactured is insulated from the remaining conductive material (such as copper). This is also called isolation milling. A small conical milling head can be used to machine an insulating channel that removes conductive material, especially copper, along the calculated movement path. The removed material may block or weaken the ability of the milling head 2 to move, and these anomalies or abnormal phenomena are identified or detected early with the help of the method according to the present invention. In one possible embodiment, corresponding evaluation numbers can also be calculated for various anomalies A, which indicate the degree of the identified anomaly or the resulting failure probability. In one possible embodiment, such an abnormal key diagram can be output to the operator of the milling machine 1 in real time via a display unit. With the method according to the present invention, possible anomalies around the clock can be identified during the ongoing milling process, and the possible causes of the error can be determined. As a result, repair and maintenance measures can be initiated or carried out promptly on the milling machine 1 concerned, thereby avoiding downtimes in the milling process to a large extent. The method according to the invention can be used for a large number of milling machines used simultaneously in a production line.

[0052] exist Figure 2 In the embodiment shown, the machine learning module 4 is implemented locally in the computing unit of the milling machine 1. In another embodiment, the machine learning module 4 can also be implemented on a remote server of a cloud platform 6, which communicates unidirectionally or bidirectionally with the controller of the milling machine 1 via a data connection. In one possible embodiment, the technical data connection includes a local network or an intranet. Alternatively, the data connection can also be a global network or the Internet. The method for monitoring the milling process according to the invention is a computer-implemented method, which is stored on a computer-readable medium in one possible embodiment. The computer-readable medium includes, for example, a data memory of the milling machine 1. In addition, the computer-readable medium can also have a portable data carrier, such as a CD-ROM or a USB stick. Depending on the application, different adaptive algorithms can be used in one possible embodiment to detect anomalies in the milling process. For example, a support vector machine can be provided to calculate quality indicators. In one possible embodiment, additional sensors are provided near the milling head 2, which provide additional data to the artificial intelligence module 4. For example, an image sensor can provide image data to the artificial intelligence module 4 as an input operating parameter I of the state vector z. As a result, the reliability or accuracy of the method according to the invention can be further improved.

[0053] Figure 3A further exemplary embodiment for monitoring a milling machine during operation, i.e. the milling process, is schematically shown. In the exemplary embodiment shown, the milling machine 1 provides operating parameters recorded via an interface via a data network or a cloud 5 to a cloud platform 6 in which an artificial intelligence module 4 is integrated. Anomalies or information can be output to the operator of the milling machine 1 in real time via a display device 7. In one possible embodiment, the display device 7 is also connected to the data network or the cloud 5. In a possible embodiment variant, the display device 7 is a portable device, such as a smartphone belonging to the user. With the method according to the invention, deviations in the feature space of the operating parameters are detected at an early stage. With the aid of the machine learning model M, the cause of anomalies or malfunctions in the milling machine 1 can also be automatically determined. Figure 3 In the exemplary embodiment shown, two characteristics or operating parameters, namely the supply current I and the rotational speed R, are used to train the machine learning model. In one possible embodiment, the trained machine learning model requires a relatively small storage space of less than 10 MB. This makes it possible to implement the adaptive algorithm from the cloud platform 6 on the automation device. For a large number of milling machines 1, the method according to the invention can be scaled in a simple manner. In one possible embodiment, the machine learning module 4 provides an evaluation number for the identified anomaly or possible failure probability. In a possible embodiment of the method according to the invention, the operating parameter P is evaluated on the server of the cloud platform 6 using the trained adaptive algorithm based on operating data sent from one or more milling machines via the network 5 to the server of the cloud platform 6. In one possible embodiment, the adaptive algorithm can be downloaded as an application from a database and executed by the artificial intelligence module 4 after the training has taken place. In one possible embodiment, the machine learning module 4 can be provided locally in the milling machine 1, such as Figure 2 As shown, or remotely implemented on a server of the cloud platform 6, such as Figure 3 shown.

[0054] Go to Figure 4 , a production line comprising a plurality of milling machines 1a, 1b, 1c is shown. The milling machines 1a, 1b, 1c may be connected to a gateway or other type of device providing a connection to a cloud platform. A machine learning model may be associated with each milling machine 1a, 1b, 1c. Since each milling machine is used at a different station of the production line, the tasks and operating modes as well as the workpieces produced by the milling machines may be different. Furthermore, the operator of the production may install milling machines of different manufacturers on the production line. Therefore, the milling machines may differ in characteristic operating set points, for example, in particular in terms of rotation speed and current.

[0055] Figure 5 The different operating spaces of different milling machines are shown. Figure 5Scatter plots of rotational speed (x-axis) and current (y-axis) for different types of milling spindles and / or milling machines (highlighted in color) are shown. The red, orange and yellow scatter plots show anomalies, while the blue scatter plot has no anomalies. The anomaly is reflected in the large values ​​of the current. This is due to the motor trying to keep the milling frequency constant during operation. However, if the milling dust jams the spindle, the current needs to be increased to keep the frequency or rotational speed constant. It can be easily seen that the anomaly of one spindle is not necessarily similar or on the same scale as the anomalies of other spindles. Different spindles operate on different rotational speed and current scales. To this end, after the machine learning model is deployed, the model is trained or fine-tuned so as to automatically learn the scale on which the spindle operates.

[0056] Now go to Figure 6 Before fine tuning, the data is pre-processed to remove the spindle acceleration and deceleration phases. These phases occur when the milling axis starts (acceleration) or stops (deceleration). After removal, the current and speed should be fairly constant during the milling process, since the spindle runs at a constant current and speed. The remaining pre-processed data is in Figure 6 The different operating modes of the spindle can thus be identified. For example, the spindle corresponding to the blue scatter plot has two different operating modes: one at 36k rpm and the other at 40k rpm.

[0057] The preprocessed data is then fed into a machine learning model, which can run in at least two phases: a calibration phase and a prediction phase.

[0058] During the calibration phase, a machine learning model can be trained. The machine learning model can be a one-step-ahead predictor, a spatial detector, or Threshold.

[0059] Figure 7 Two different phases are shown. During the calibration phase, the first time series data is obtained and a model, such as an autoregressive moving average model (ARIMA) of categories, is trained on the data in interval I. The model is then used in interval II to compare the predictions made, for example, by a one-step-ahead predictor (green dots) with the observed data (white dots). This results in a distribution over the residuals (differences between observations and predictions) and defines the domain of residuals for normal operation. A very precise fit or machine learning model is not required for the data in interval I. The only requirement is that the machine learning model is able to make better predictions for normal data than for abnormal data.

[0060] In interval III (prediction phase), the residuals are calculated and compared with the distribution of interval II. If the distance to the median or mean of the distribution is too large (e.g., in terms of interquartile range or standard deviation), the corresponding data point can be identified as an anomaly.

[0061] More details of the machine learning model, especially the one-step-ahead predictor, can be found in Environmental Modeling and Software 25 (2010) 1014-1022 / DJ Hill, BS Minsk "Anomaly Detection in Streaming Environmental Sensor Data: A Data-Driven Modeling Approach".

[0062] The same (two-stage) approach can be used for a machine learning model in the form of a spatial detector, where, for example, a class of support vector machines is trained on time series data for interval I, including time series data for current I and rotation speed S. An anomaly score can be calculated based on the distribution of the time series data during interval II, for example by calculating the distance to the hyperplane of the support vectors. The anomaly score calculated during the prediction phase (interval III) can then be compared to the distribution.

[0063] Alternatively, you can use Threshold model. In this case, the threshold can be calculated using the data of two time series data of intervals, such as the time series data of intervals I and II. Subsequently, the new time series data (such as interval III) can be compared with the threshold.

[0064] Depending on the specific underlying model, intervals I, II, and III may also be moved forward in time, yielding a type of continuous learning, e.g., as described above. However, the model and time series data may need to meet one or more of the following requirements:

[0065] 1. No overfitting using a small dataset (in order to shorten the duration of the calibration phase; about 24 hours).

[0066] 2. Fast prediction (low latency).

[0067] 3. Computationally cheap (does not require large amounts of memory and CPU; can be easily run on edge devices).

[0068] 4. Intervals I and II should not contain any (or only a few) anomalies.

[0069] In addition, the prediction phase (interval III) can be divided into multiple windows. For example, in one or more large windows L and multiple small windows S. The score distribution of the small window can then be compared with the large window. This produces a model that automatically updates the distribution (of interval II), but no retraining is required. In other words, interval I is no longer required, but it still allows local anomalies (i.e., in small window S) to be discovered by comparing the anomaly score with the larger background (i.e., window L).

[0070] Furthermore, an aggregation of the models used is proposed. Here, majority voting is proposed. In which, if a majority of detectors (e.g., a majority) identify / label a data point as an anomaly, then the data point is fully identified as an anomaly.

[0071] Finally, hysteresis can be implemented, i.e. an alert is sent to the client (only) if a certain number of anomalies occur within a defined time frame, in order to reduce the number of false alerts.

[0072] Typically, an untrained machine learning model can be packaged as an app in an app store, e.g. on the aforementioned cloud platform, which customers can download and apply to their respective (milling) machines. During a calibration phase that can be started automatically, e.g. after downloading, the parameters for the anomaly detection model are learned. Furthermore, the proposed method can be implemented as a service delivery at the customer machine. Furthermore, a cloud platform such as MindSphere can provide further service and maintenance services for a fleet of milling machines.

[0073] Steering Figure 8 , further exemplary method steps are shown. In a first step S1, it is proposed to employ an untrained machine learning model to determine one or more anomalies in the time series data. In this document, untrained shall mean uninitialized. Untrained may also be understood to refer to a machine learning model whose variables / parameters are initialized at deployment, e.g., by assigning random values ​​to the variables / parameters, but are later overwritten during training or by training the machine learning model.

[0074] For example, the program code (e.g., Python) in which the machine learning model is written may include variables or parameters to be trained, where the variables or parameters are not initialized in the untrained machine learning model. In this case, the variables / parameters are still unknown and during training, i.e., during the calibration phase, for example, Figure 7 The machine learning model can be used to observe the spindle of the milling machine during the training period, for example during the first hour of operation of the milling machine, and derive variables / parameters therefrom. Only then can the machine learning model be used to "predict" or calculate anomaly scores.

[0075] The simpler the machine learning model, the lower the number of hours or amount of first time series data required. For this purpose, preferably, the first time series data includes different operating states, preferably all operating states of the milling machine, in order to learn the (nominal) behavior of the milling machine. This is because different milling machines, especially different spindles, may have different operating states. For this purpose, it has been found that the first time series data optimally covers a period of two weeks.

[0076] Thus, in a second step S2, during operation of the milling machine, first time series data representing a rotational speed of a milling head of the milling machine and at least one further operating parameter of the milling machine may be obtained by means of an untrained machine learning model. As described above, the power supply current for the milling machine or the milling process is controlled so as to operate at a constant rotational speed, so that if the spindle is stuck, for example in the case of one or more chips, the power supply current is increased.

[0077] As mentioned above, the first time series data may be obtained during a first or initial operating time of the milling machine - assuming that operation of the machine during this time is nominal.

[0078] In a third step S3, an untrained machine learning model is trained during operation of the milling machine based on the obtained first time series data. This allows, for example, existing milling machine and production line installations to be enhanced by the proposed monitoring method without interrupting operation.

[0079] In a fourth step S4 , the trained machine learning model obtains second time series data representing the rotation speed of a milling head of the milling machine and further operating parameters, such as power supply current, during operation of the milling machine.

[0080] The first and / or second time series data may be obtained from a CPU, PLC and / or controller of the milling machine, a converter powering the milling machine and / or from a sensor element connected to the milling machine.

[0081] In a fifth step S5, the trained machine learning model determines one or more anomalies in the second time series data during operation of the milling machine.

[0082] Go to Fig. 9 , other steps of the exemplary method are shown.

[0083] In step S6, an acceleration phase and / or a deceleration phase of the milling head is determined. The acceleration phase and / or the deceleration phase comprises one or more data points in the first time series data below a first threshold value of the rotation speed of the milling head.

[0084] In step S7, the determined data points of the acceleration phase and / or the deceleration phase are removed from the first time series data. In addition, corresponding data points of other operating parameters can be removed from the first time series data, preferably, other data points within a predetermined distance from one or more determined data points are also removed.

[0085] In step S8, the untrained machine learning model is then trained based on the remaining data points in the first time series data.

[0086] Furthermore, data points in the first time series data between a deceleration phase and a subsequent acceleration phase of the milling head are optionally removed.

[0087] It should be understood that steps S6, S7 and S8 may be performed in conjunction with steps S1 to S5, preferably between steps S2 and S3 and / or between steps S4 and S5, or may be performed independently, i.e. historical data may be processed as described in steps S6 to S8.

[0088] Go to Fig.10 , other steps of the exemplary method are shown.

[0089] In step S9, one or more data points are predicted based on the first subset of the first time series data, preferably based on the ARIMA model. The prediction may be part of the training of an untrained machine learning model. To this end, one or more data points of the first time series data are used to predict future data points in the time series.

[0090] In step S10, the first deviation between the predicted data points and the data points in the second subset of the first time series data is calculated. To this end, the first time series data is divided into a first subset used as a basis for prediction and a second subset for comparison of predicted values ​​with actual values. The first subset is used to fit a forecasting model, such as ARIMA, and the second subset determines which residuals are "normal" and makes a simple "comparison" during the forecasting phase (including the second time series data), for example, how much of the interquartile range is the new residual away from the median.

[0091] Now, in step S11, a first threshold is determined based on the first deviation. The first threshold can then be compared with the data points of the second time series data. Thus, for example, during operation of a milling machine, an anomaly can be determined based on this comparison, since the threshold serves as a measure for determining whether the second time series data is characteristic of a rated behavior. A previously untrained machine learning model is then trained.

[0092] Alternatively, in step S12, a probability distribution of deviations between the predicted data points and the data points in the second subset of the first time series data can be determined. Thus, the distribution of deviations from our forecast / predicted values ​​is determined. This has the advantage that anomalies can be detected that not only contain the absolute value of the deviation of the data, but also that the deviation from the normal operating point is taken into account when determining the anomaly, which is individual for each milling machine. A previously untrained machine learning model is then trained.

[0093] It should be understood that steps S9 to S11 or S9 to S12 may be combined with Figure 8 and / or Fig. 9 The method steps may be performed in combination or may be performed independently.

[0094] Go to Fig.11In step S13, data points in the second time series data corresponding to the acceleration and / or deceleration of the milling head are removed. Additionally or alternatively, data points in the second time series data between the deceleration and the successive acceleration of the milling head are removed. Thus, a data set equivalent to the first time series data is obtained, i.e., the second time series data.

[0095] Go to Fig.12 In step S15, one or more data points may be predicted based on, for example, the first and / or second subset of the first time series data, preferably using ARIMA.

[0096] Now, the second time series data may include data representing later behavior during the life of the milling machine, for example, the second time series data may represent wear or other abnormal behavior of the milling head.

[0097] In step S16 , a second deviation between the predicted data point and the data point in the second time series data may be determined.

[0098] In step S17, the second deviation is compared with a dispersion measure of the probability distribution, such as the interquartile range.As mentioned above, the proposed method and embodiments allow monitoring of milling machines, each having its own operating characteristics.

[0099] In a subsequent step S21, an anomaly is detected if the second deviation exceeds the deviation measure. The detected anomaly can be output. In addition, a counter can be increased in the event of a detected anomaly. Based on the counter, a remaining service life or an external maintenance measure can be initiated.

[0100] Go to Fig.13 , in step S18, the second time series data is divided into a plurality of subsets, and a third deviation between the predicted data points and the data points in the subset is determined for each subset, and in step S19, the corresponding third deviation of the subset is compared with a deviation measure of the probability distribution, and in step S20, the probability distribution is updated based on the comparison.

[0101] Now go to Fig.14 In step S22, a support vector machine SVM is trained based on one or more data points of a first subset of the first time series data. In step S23, a first deviation between one or more hyperplanes of the SVM and data points in a second subset of the first time series data is determined.

[0102] In step S24 , a second threshold based on the first deviation is determined, wherein the second threshold is used to compare the data points of the second time series data with the second threshold.

[0103] Alternatively, in step S25, a probability distribution of first deviations between one or more hyperplanes and data points in the second subset of the first time series data is determined. A previously untrained machine learning model is then trained.

[0104] Go to Fig.15 In step S26, a second deviation between one or more hyperplanes of the SVM and the data points in the second time series data is determined. In a subsequent step S27, the second deviation is compared to a dispersion measure, such as the interquartile range of the probability distribution. In step S28, an anomaly is identified if the second deviation exceeds the dispersion measure.

[0105] Go to Fig.16 In step S29, according to at least two different methods, for example, according to Fig.12 and Fig.15 In an embodiment, anomaly candidates, i.e., one or more data points, are determined. Anomalies are identified only when two methods produce the same results. In the case of using more than two methods, the majority of the results can then determine whether these anomaly candidates are identified as anomalies.

Claims

1. A method for monitoring a milling machine (1), comprising the following steps: - deploying (S1) an untrained machine learning model (M) for determining one or more anomalies in the time series data; - during operation of the milling machine (1), obtaining (S2) by the untrained machine learning model (M) from a converter supplying power to the milling machine (1) first time series data (I, II) representing a rotation speed of a milling head of the milling machine (1) and at least one further operating parameter of the milling machine (1); as well as - during operation of the milling machine (1), training (S3) the untrained machine learning model (M) based on the first time series data (I, II) obtained, - during operation of the milling machine, obtaining (S4) second time series data (III) representing the rotation speed of the milling head and further operating parameters of the milling machine (1) by the trained machine learning model (M), - during operation of the milling machine (1), determining (S5) one or more anomalies in the second time series data (III) by the trained machine learning model (M), The method further comprises the following steps: - determining an acceleration phase and / or a deceleration phase of the milling head, the acceleration phase and / or the deceleration phase comprising one or more data points in the first time series data (I, II) below a first threshold value of the rotation speed of the milling head, - removing the determined data points of the acceleration phase and / or deceleration phase and the corresponding data points of the further operating parameters from the first time series data (I, II), and also removing further data points within a predetermined distance from one or more determined data points, - training the untrained machine learning model based on the remaining data points in the first time series data (I, II).

2. The method according to claim 1, comprising the steps of: - removing data points in the first time series (I, II) of data between a deceleration phase and a subsequent acceleration phase of the milling head.

3. The method according to claim 1, wherein: Training the untrained machine learning model (M) based on the first time series data comprises: - predicting one or more data points based on a first subset (I) of the first time series data (I, II) using an ARIMA model.

4. The method according to claim 3, wherein: Training the untrained machine learning model (M) based on the first time series data comprises: - determining a first deviation between the predicted one or more data points and one or more data points in a second subset (II) of the first time series data (I, II).

5. The method according to claim 4, wherein: Training the untrained machine learning model (M) based on the first time series data includes: determining a first threshold based on the first deviation, wherein the first threshold is used to compare data points of the second time series data (III) with the first threshold.

6. The method according to claim 3, wherein: Training the untrained machine learning model (M) based on the first time series data comprises: - determining a probability distribution of deviations between the predicted data points and data points in a second subset (II) of the first time series (I, II) data.

7. The method according to claim 1, wherein: During the operation of the milling machine (1), determining, by the trained machine learning model, one or more anomalies based on the second time series data (III) comprises: - removing data points corresponding to acceleration and / or deceleration of the milling head in the second time series data (III), and / or - removing data points between the deceleration and the subsequent acceleration of the milling head in the second time series data (III).

8. The method according to claim 6, wherein: Determining, by the trained machine learning model (M), one or more anomalies based on the second time series data during operation of the milling machine comprises: - using ARIMA, forecasting one or more data points based on the first subset (I) and / or the second subset (II) of the first time series data (I, II), - Determining a second deviation between the predicted data point and the data points in the second time series data (III).

9. The method according to claim 8, wherein: Determining, by the trained machine learning model, one or more anomalies based on the second time series data during operation of the milling machine comprises: - comparing the second deviation to a dispersion measure of the probability distribution.

10. The method according to claim 6, further comprising: - dividing the second time series data (III) into a plurality of subsets and determining for each subset a third deviation between the predicted data points and the data points within the subset, and - comparing the respective third deviations of the subsets to a measure of dispersion of the probability distribution, and - updating the probability distribution based on the comparison.

11. The method according to claim 9, wherein: Determining, by the trained machine learning model (M), one or more anomalies based on the second time series data during operation of the milling machine comprises: - identifying an anomaly if the second deviation exceeds the dispersion measure.

12. The method according to claim 6, wherein: Based on the first time series data (I, II), training the untrained machine learning model (M) comprises: - training a support vector machine SVM based on one or more data points of a first subset (I) of the first time series data, - determining a first deviation between one or more hyperplanes of the SVM and data points in a second subset (II) of the first time series data (I, II).

13. The method according to claim 12, wherein: Based on the first time series data (I, II), training the untrained machine learning model (M) includes: determining a second threshold based on the first deviation, and comparing data points of the second time series data with the second threshold.

14. The method according to claim 12, wherein: Based on the first time series data, training the untrained machine learning model comprises: - determining a probability distribution of the first deviations between the one or more hyperplanes and data points in the second subset (II) of the first time series data (I, II).

15. The method according to claim 12, wherein: Determining, by the trained machine learning model, one or more anomalies based on the second time series data during operation of the milling machine comprises: - determining second deviations (III) between the one or more hyperplanes of the SVM and data points in the second time series data.

16. The method according to claim 15, wherein: Determining, by the trained machine learning model, one or more anomalies based on the second time series data during operation of the milling machine (1) comprises: - comparing the second deviation with a dispersion measure of the probability distribution.

17. The method according to claim 6, further comprising: - dividing the second time series data (III) into a plurality of subsets, and determining for each subset a third deviation between the predicted data point and the data points in the subset, and - comparing the respective third deviations of the subsets to a measure of dispersion of the probability distribution, and - updating the probability distribution based on the comparison.

18. The method according to claim 16, wherein: Determining, by the trained machine learning model, one or more anomalies based on the second time series data (III) during operation of the milling machine (1) comprises: - identifying an anomaly if the second deviation exceeds the dispersion measure.

19. A computer program product comprising program code, which when executed performs the steps of the method of any one of claims 1 to 18.

20. A device for monitoring a milling machine comprising a processor and a memory, the device being operated to perform the steps of the method according to any one of claims 1 to 18.

Images