Computer-implemented method and condition monitoring device for detecting mechanical damage of machine

Abstract

A computer-implemented method for identifying mechanical damage to a machine on the basis of performance indicators present at a rotational speed of the machine, comprising the following steps:-receiving, during a calibration period, a plurality of calibration data points, each comprising a rotational speed of the machine and a performance characteristic value assigned to the rotational speed, and is determined from sensor data of a machine operating at one or more working points; -calibrating the AI-based Gaussian process regressor by inputting calibration data points received during a calibration period and outputting the estimated performance characteristic curve and a range of uncertainty assigned to each value thereof for all rotational speed values within a specified rotational speed range indicative of a normal state of the machine; receiving the data point to be evaluated during an evaluation period following the calibration period; and-comparing it with the estimated performance characteristic curve; and outputting an exception message if it lies outside at least one specified limit value dependent on the range of uncertainty of the performance characteristic value.

Description

Technical Field

[0001] The present invention relates to a computer-implemented method for identifying mechanical damage of a machine based on performance indicators present at the rotational speed of the machine, a condition monitoring device for executing the method, and a corresponding computer program product. Background Art

[0002] Condition monitoring systems are used to monitor the condition of machinery and equipment and identify potential problems early. Based on historical sensor data, such as temperature or vibration, data analytics methods are used to detect deviations from "normal conditions" and take action when necessary. Early identification of these deviations allows maintenance cycles to be optimized and unplanned downtime to be avoided.

[0003] A major challenge here is learning this "normal state" for the machine. Furthermore, rapid commissioning is required: the condition monitoring system must be able to detect deviations from the machine's normal state after the shortest possible calibration time. However, from a system perspective, this requires sufficient data to learn the "normal state" across the machine's various states.

[0004] When monitoring the condition of machines, especially those used in industrial applications, vibration is a key measurement parameter for identifying machine faults such as bearing damage, shaft misalignment, or imbalance. In practice, various performance indicators (KPIs) are often derived from high-resolution vibration data. These performance indicators are then observed over time, and trend and anomaly detection algorithms are applied to these time-dependent performance curves.

[0005] However, electric motors, particularly those operated with converters in different speed ranges, exhibit very different vibration behaviors in different speed ranges, particularly due to resonance effects. This makes the resulting KPIs difficult to compare. Therefore, when analyzing performance indicators derived from vibration data, the motor's operating point must be considered. However, this requires that the condition monitoring system be provided with sufficient historical data from different motor operating points during production operation to allow for interpretation of possible anomalies within a specific speed range. This cannot be guaranteed, particularly in the immediate period after motor commissioning. Summary of the Invention

[0006] The present invention is therefore based on the task of providing a prediction for a performance indicator even after a short calibration time, i.e. based on a small amount of historical data, and in a reliable manner. Another task is to determine a prediction for a performance indicator at previously unobserved or barely observed machine operating points.

[0007] This object is achieved by the measures described in the independent claim. Advantageous developments of the invention are disclosed in the dependent claims.

[0008] A first aspect of the present invention relates to a computer-implemented method for identifying mechanical damage of a machine based on a performance indicator related to the rotational speed of the machine, the method comprising the steps of:

[0009] - receiving, during a calibration period, a plurality of calibration data points, each of which includes a rotational speed of the machine and a performance characteristic value assigned to the rotational speed, and wherein the calibration data points are determined based on sensor data of the machine operating at one or more operating points;

[0010] - calibrating the AI-based Gaussian process regressor by inputting calibration data points received during a calibration period, and outputting an estimated performance characteristic curve and uncertainty ranges assigned to each value of the performance characteristic curve for all speed values ​​within a specified speed range indicative of a normal state of the machine;

[0011] - receiving data points to be evaluated during an evaluation period following a calibration period; and

[0012] - comparing the data point to an estimated performance characteristic curve; and

[0013] If the data point to be evaluated lies outside at least one specified limit value of the uncertainty range dependent on the performance characteristic value, an exception message is output.

[0014] Advantageously, the Gaussian process regressor not only predicts the performance characteristic curve but also models the uncertainty of each performance characteristic value. This uncertainty range significantly reduces the calibration time of such a system, as it eliminates the need for sufficient data points for all operating points. Each data point—that is, each calibration data point and each data point to be evaluated—includes a speed and the performance characteristic value determined at that speed. This uncertainty range allows for an estimate of the plausibility of each evaluated data point.

[0015] In one advantageous embodiment, the Gaussian process regressor includes a statistical model for regression and prediction of data points, the statistical model including a covariance function for modeling a relationship between input values ​​and output values, wherein hyperparameters of the covariance function are set by the calibration. The performance characteristic curve is a mean value determined by the Gaussian process regressor, and the uncertainty range is a standard deviation determined by the Gaussian process regressor.

[0016] In one advantageous embodiment, the distance of the data point to be evaluated from the mean value is calculated and compared with the standard deviation of the data point on the performance characteristic curve.

[0017] In other words, for each data point to be evaluated, i.e., the performance characteristic value determined at this speed, the distance to the estimated performance characteristic value on the estimated performance characteristic curve is determined, and this distance is compared with the uncertainty range. This prevents erroneous exception messages for data points to be evaluated at speeds with large uncertainty ranges.

[0018] In an advantageous embodiment, the performance characteristic curve comprises characteristic performance values ​​for all rotational speed values ​​within a specified rotational speed range.

[0019] The performance characteristic curve can therefore be used for all data points to be evaluated whose speed lies within the speed range.

[0020] In an advantageous embodiment, the Gaussian process regressor is recalibrated with additional data points at specified time intervals.

[0021] Only the performance characteristic curve, and therefore the Gaussian process regressor, needs to be recalibrated to improve accuracy. The method is thus optimized in terms of processor power. By outsourcing the calibration process to a remote server, the data points to be evaluated can be evaluated by a simple device located near the machine.

[0022] Advantageously, during the calibration period, an increased number of data points per unit time is received relative to the evaluation period.

[0023] Therefore, the calibration period can be shortened or the amount of calibration data used for the recalibration can be increased, and thus the calibration result can be improved, ie, a more accurate performance characteristic curve can be output.

[0024] Preferably, data points received during a previous operating period of the machine are used as new calibration data points.

[0025] In one advantageous embodiment, the rotational speed is derived from the magnetic field data, and the performance characteristic value is derived from vibration data measured at the machine.

[0026] Thus, a physically reliable relationship is given between the value of the data point and the value of the machine parameter measured directly at the machine.

[0027] In an advantageous embodiment, a deterioration message is output if the number of abnormal messages received for a data point within a specified time period exceeds a specified limit value.

[0028] By means of the deterioration message, a deterioration of the state of the machine can be actively indicated. In the case of a plurality of specified, in particular increased, limit values, different criticality levels can be implemented for the machine.

[0029] In an advantageous embodiment, the estimated performance characteristic curve and the uncertainty range assigned to each value of the performance characteristic curve as well as the data points to be evaluated are output on a graphical user interface.

[0030] By displaying the status of the machine and its behavior over the entire speed range and optionally over time, maintenance personnel receive early and comprehensive indication of impending hazardous situations or maintenance work.

[0031] In an advantageous embodiment, the mechanical damage is a bearing damage or a misalignment of a shaft of the machine or an imbalance of the machine.

[0032] Through exception messages, the most common causes of degradation of machine operation can be identified.

[0033] According to a second aspect, the present invention relates to a condition monitoring device for identifying mechanical damage of a machine based on performance indicators present at a rotational speed of the machine, the condition monitoring device comprising at least one processor configured to execute the following steps:

[0034] - receiving, during a calibration period, a plurality of calibration data points, each of which includes a rotational speed of the machine and a performance characteristic value assigned to the rotational speed, and wherein the calibration data points are determined based on sensor data of the machine operating at one or more operating points;

[0035] - calibrating the AI-based Gaussian process regressor by inputting calibration data points received during a calibration period, and outputting an estimated performance characteristic curve and uncertainty ranges assigned to each value of the performance characteristic curve for all speed values ​​within a specified speed range indicative of a normal state of the machine;

[0036] - receiving data points to be evaluated during an evaluation period following a calibration period; and

[0037] - comparing the data point to an estimated performance characteristic curve; and

[0038] If the data point to be evaluated lies outside at least one specified limit value of the uncertainty range dependent on the performance characteristic value, an exception message is output.

[0039] The described condition monitoring device is able to perform robust anomaly detection using relatively few data points and to model the speed dependence of performance indicators (also referred to as KPIs) of machine damage with sufficient accuracy. Even if a rotating machine, in particular a motor, is operated in a different operating state at a later time after a calibration phase, the extent to which this is an anomaly can be determined.

[0040] According to a third aspect, the invention relates to a computer program product comprising a non-volatile computer-readable medium directly loadable into the memory of at least one digital computer, the non-volatile computer-readable medium comprising program code portions which, when executed by the at least one digital computer, cause the at least one digital computer to perform the steps of the method.

[0041] A computer program product, such as a computer program means, can be provided or supplied, for example, as a computer-readable storage medium in the form of a memory card, USB memory stick, CD-ROM, DVD or also in the form of a downloadable file from a server in a network. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Exemplary embodiments of the method according to the invention and the device according to the invention are shown by way of example in the drawings and are explained in more detail in the following description.

[0043] Figure 1 An exemplary illustration of data points recorded on a machine having two working points is shown schematically;

[0044] Figure 2 An embodiment of the method according to the present invention for detecting mechanical damage to a machine is shown in the form of a flow chart;

[0045] Figure 3 An exemplary embodiment of an estimated performance characteristic curve with uncertainty ranges determined by the method according to the present invention is shown;

[0046] Figure 4 A schematic diagram shows an exemplary process of the method according to the present invention during a calibration period and an evaluation period; and

[0047] Figure 5 An embodiment of the condition monitoring device according to the present invention is shown in a block diagram.

[0048] In the embodiments and figures, identical or identically functioning elements may be provided with the same reference numerals. In principle, the illustrated elements and their dimensional relationships to one another should not be considered to be to scale. Rather, individual elements may be illustrated on a larger scale for better presentation and / or understanding.

[0049] Unless otherwise stated in the subsequent description, the terms "receiving", "calibrating", "comparing", "outputting" etc. preferably relate to operations and / or processes and / or processing steps that change and / or generate data and / or convert these data into other data, wherein these data can in particular be presented or exist as physical quantities, for example as electrical pulses.

[0050] The condition monitoring device and the components contained therein, such as at least one data interface, a regressor unit, an evaluation unit, or an output interface, etc., may include one or more processors. The processor may be, in particular, a central processing unit (CPU), a microprocessor or a microcontroller, such as an application-specific integrated circuit or a digital signal processor, etc., which may be combined with a memory unit for storing program instructions. DETAILED DESCRIPTION

[0051] Previous condition monitoring systems either completely decoupled the performance indicators (hereinafter also synonymously referred to as KPIs) derived from vibration data or motor current data from the current operating point of the machine, or, for example, divided the possible bandwidth of the speed s into a plurality of fixed operating points BP1, BP2, such as Figure 1 However, this approach has disadvantages. If the resulting KPIs, which may, for example, measure the severity of motor damage, are not set in relation to the operating point, it is difficult to clearly identify trends or anomalies in these KPIs, as they may fluctuate significantly with the operating point.

[0052] Divided into separate, strictly separated working points BP1, BP2, such as Figure 1 The problem is only partially solved as shown in . For example, if outlier detection, also known as outlier detection, is performed based on the distributions V1, V2 of previous data points in the corresponding working points BP1, BP2, and thresholds k1, k2 or k2, k3 are defined for data points outside a certain confidence interval, then the following problems may arise:

[0053] While the dependence of the performance indicator KPI on the machine speed s is approximately linear and therefore exhibits no abnormal behavior, the two resulting distributions V1 and V2 differ significantly. For example, data point DP1, located just to the right of the dividing line between operating points BP1 and BP2, is within the normal range for operating point BP1 but is considered abnormal for operating point BP2 because it significantly exceeds distribution V2 and lies below threshold k2.

[0054] Therefore, an arbitrary division into a series of operating points BP1, BP2 cannot be performed, as this could lead to false positive detections in some cases.Since the specific vibration behavior can vary greatly from motor to motor within the speed range and cannot be known in detail in practice, it is not possible to rationally select the limit values.

[0055] Therefore, the motor's operating point must be considered when analyzing performance indicators derived from vibration data. However, this requires that, during production operation, the data-based, machine learning-based condition monitoring system be provided with sufficient historical data from different motor operating points to allow for interpretation of possible anomalies within a specific speed range. This is particularly difficult to achieve shortly after motor commissioning.

[0056] The proposed computer-implemented method for identifying mechanical damage, for example in rotating machines, solves this task. Figure 2 An exemplary process of the method is schematically shown as a flow chart.

[0057] The basic idea of ​​this method is to use a Gaussian process regressor to model performance characteristics that are used as indicators of motor mechanical damage based on vibration data and are strongly correlated with speed. Specifically, the following steps are performed:

[0058] During a calibration phase, a plurality of calibration data points kDP are received (see step S1), each comprising a rotational speed of the machine and a performance characteristic value assigned to the rotational speed. The rotational speed is determined based on sensor data of the machine operating at one or more operating points. The rotational speed is derived from magnetic field data, and the performance characteristic value is derived from vibration data measured at the machine.

[0059] In step S2, the AI-based Gaussian process regressor (GPR) is calibrated using the calibration data points (kDP) received during the calibration period. The calibrated Gaussian process regressor (GPR) provides, as output, an estimated performance characteristic curve (gKPI) and an uncertainty range assigned to each value of the performance characteristic curve for all speed values ​​within a specified speed range. The estimated performance characteristic curve (gKPI) indicates a normal state of the machine.

[0060] Calibration of the Gaussian process regressor GPR is thus completed, and the estimated performance characteristic gKPI can be used to evaluate data points recorded during machine operation. Accordingly, during an evaluation period following the calibration period, a data point bDP to be evaluated is received (see S3). This data point bDP is compared with the corresponding performance indicator of the estimated performance characteristic gKPI at the speed of this data point bDP (see S4). If the data point bDP to be evaluated lies outside at least one specified limit value, an exception message AM is output, where the at least one limit value depends on the uncertainty range of the performance characteristic value (see S5).

[0061] Preferably, an exception message AM is output if the evaluated data point bDP lies outside the uncertainty range, i.e., if the characteristic value of the performance of the data point bDP is greater than the upper limit of the uncertainty range, the maximum characteristic value, or if the characteristic value of the performance of the data point is less than the lower limit of the uncertainty range, the minimum characteristic value. The limit value may also be specified within and / or outside the uncertainty range.

[0062] If the data point bDP to be evaluated is within the specified limit values, no abnormality message is output, see step S6. In this case, a "normal status message" can optionally be output, which confirms that the machine is in a normal state within the specified limit values. This can be indicated, for example, by outputting the evaluated data point in a performance characteristic curve diagram, such as Figure 3 As shown in .

[0063] The Gaussian process regressor (GDP) includes a statistical model for regression and prediction of data points. It uses a covariance function to model the relationship between input and output values. During calibration, the hyperparameters of the covariance function are estimated based on the received calibration data points kDP to fit the regressor. Using this model, predictions for new input values ​​are calculated by generating a probability distribution over possible output values. The GDP also provides uncertainty estimates for its predictions.

[0064] In the concrete case, the model primarily provides a continuous function that describes the dependence of the characteristic performance values ​​on the rotational speed within a specified rotational speed range and thus represents the estimated characteristic performance curve. Furthermore, the model provides information on the uncertainty within this rotational speed or rotational speed range.

[0065] Figure 3 An example of an estimated performance characteristic curve 11 over a specified speed range is shown, the estimated performance characteristic curve being representative of normal operation of the machine and being output by a Gaussian process regressor based on approximately ten received calibration data points 12 .

[0066] Graph 10 shows this performance characteristic curve 11, which represents the dependency between the performance characteristic value and the speed s, modeled using a Gaussian process regressor. The shaded area shows a confidence interval, for example a 90% confidence interval, which indicates the uncertainty range 13. This means that there is a 90% probability that uncertainty range 13 contains the true value of the performance indicator. It can be seen that this dependency is not linear. While a linear increase is observed in the speed range between 2600 and 2700 rpm, the performance indicator decreases again with increasing speed around 2800 rpm. Due to the small divergence here, the uncertainty of this method is very low, making it sensitive to outliers. However, the uncertainty increases sharply in the range below 2400 rpm and above 2900 rpm, as there are no calibration data points from which a "normal state" can be reliably derived. Nevertheless, a rough statement can still be made, since a continuous function representing the performance characteristic 11 is learned based on previous calibration data points 12 , which describes it as well as possible and taking into account the uncertainties of the method.

[0067] Since this uncertainty can be modeled using a Gaussian process, the calibration time for this method and the condition monitoring device implementing it can be significantly reduced, as sufficient calibration data points are not absolutely necessary for all operating points of the machine. At the same time, the method is robust against false positive anomaly detections and corresponding anomaly messages.

[0068] Preferably, the Gaussian process regressor is recalibrated at specified time intervals using additional data points recorded on the machine at later points in time, thereby improving the accuracy of the resulting, newly estimated performance characteristic curve and uncertainty range.

[0069] Figure 4 An embodiment of the method is shown with details regarding the recalibration of the Gaussian process regressor and the recalibrated estimated performance characteristic curve derived therefrom.

[0070] By means of a sensor device on the machine to be monitored, brief, high-resolution snapshots of vibration data VD and magnetic field data FD are recorded at intervals during a calibration period, and optionally other data used to calculate the rotational speed of the machine (see M1, M2). In these high-resolution snapshots, a large amount of vibration data VD and magnetic field data FD is measured. These time intervals can be predefined cyclically or according to any other scheme, or can be triggered by one or more events.

[0071] The current rotational speed s of the machine at the time of recording is then calculated from the magnetic field data FD (see M3) and stored (see M5). Performance characteristic values ​​KPI are calculated from the vibration data VD using various signal processing methods (see M4) and likewise stored, for example in a database (see M5).

[0072] Two values ​​are coupled as a data point kDP, namely the speed s and the associated performance characteristic KPI, and serve as input for calibrating a Gaussian process regressor or regression model (see M6). This Gaussian process regressor or regression model calculates a Gaussian distribution for a given speed s, which describes the "normal behavior" of the machine within this speed range via its mean and standard deviation. The calibrated Gaussian process regressor GPR now provides the performance characteristic curve for condition monitoring (see M7).

[0073] Here, an initial number N of data points is used to essentially calibrate the regression model. Data points recorded at later points in time are then used for retraining. Data points from value ranges for which only a few data points are available are preferably used for retraining / recalibration. For example, during operation, data from a very high speed range is added, for which only a few data points are available and for which the machine's behavior could only be poorly predicted. The resulting performance characteristic curve includes performance characteristic values ​​for all speed values ​​within the specified speed range.

[0074] Then, during the machine's condition monitoring (this is referred to as the evaluation period), see M8 , the new data point bDP to be evaluated is compared with the average value within the range. To this end, the distance between the new data point bDP to be evaluated and the average value is calculated and compared with the uncertainty range for that point on the performance characteristic curve. If the distance is greater than a specified limit, such as n times the standard deviation, the new data point to be evaluated is marked as abnormal. An abnormality message is output. Preferably, during the calibration period, an increased number of data points are received per unit time relative to the evaluation period.

[0075] The data points to be evaluated can also be used as calibration data points.

[0076] As mechanical damage progresses, an upward trend in the performance characteristic values ​​for the data points being evaluated can be expected. That is, new data points being evaluated will drift toward higher performance characteristic values. Because the model parameters of the Gaussian process regressor do not change for a certain period of time after recalibration, the distance between the new performance characteristic values ​​and the learned mean value will become increasingly larger, and after a period of time, it will exceed the previously defined limit of n times the standard deviation. This trend, and thus the worsening of the mechanical failure, can be detected by an increase in detected anomalies and corresponding anomaly messages. In one variant, in this case, the method triggers a warning and outputs a deterioration message.

[0077] Figure 5 A condition monitoring device 30 is shown for identifying mechanical damage to a machine 20 based on performance indicators present at the rotational speed of the machine 20. The condition monitoring device 30 includes at least one processor configured as: a calibration data interface 31; a regressor unit 32; a measurement data interface 33; an evaluation unit 34; and an output interface 35.

[0078] Machine 20 is a rotating machine, such as a motor, particularly an electric motor. Sensors arranged on machine 20 detect field strength and vibrations of the machine. The sensor data derives the machine's rotational speed and one or more performance indicators that are indicative of mechanical damage. Mechanical damage that can be identified by these performance indicators includes, for example, bearing damage, misalignment of machine shafts, and / or machine imbalance.

[0079] Calibration data interface 31 is configured to receive, during a calibration period, a plurality of calibration data points, each comprising a rotational speed of machine 20 and a performance characteristic value assigned to that rotational speed, wherein the calibration data points are determined based on sensor data of the machine operating at one or more operating points. Optionally, the rotational speed and the performance indicator present at that rotational speed can be derived from the sensor data within condition monitoring device 30.

[0080] The regressor unit 32 is designed to: calibrate the AI-based Gaussian process regressor by inputting calibration data points received during a calibration period; and output an estimated performance characteristic curve and an uncertainty range assigned to each value of the performance characteristic curve for all speed values ​​within a specified speed range indicating a normal state of the machine.

[0081] The measurement data interface 33 is designed to evaluate the data points to be evaluated during an evaluation period following the calibration period.

[0082] The evaluation unit 34 is designed to compare the data point to be evaluated with the estimated performance characteristic curve.

[0083] Output interface 35 is configured to output an exception message if a data point to be evaluated lies outside at least one specified limit value of an uncertainty range dependent on the performance characteristic value. Preferably, output interface 35 is configured as a graphical user interface. The graphical user interface outputs the estimated performance characteristic curve, the uncertainty range assigned to each value of the performance characteristic curve, and the data point to be evaluated.

[0084] A computer program product comprises a non-volatile computer-readable medium which can be directly loaded into the memory of at least one digital computer and which includes program code portions which, when executed by the at least one digital computer, cause the at least one digital computer to perform the steps of the above method and to operate as a condition monitoring device.

[0085] Compared to previous solutions, the described condition monitoring device 30 is able to perform robust anomaly detection using relatively few data points and model the speed dependency of the machine damage KPI with sufficient accuracy. Even if the motor is operated in a different operating state at a later time after the calibration phase, the extent to which this is an anomaly can be determined. As the number of data points increases, the behavior of the machine 20 can be modeled in increasing detail over time, by regularly retraining the Gaussian process regressor. This approach significantly reduces the calibration time of the condition monitoring system without significantly increasing the risk of false detections. This significantly improves the performance capabilities of the condition monitoring system and, consequently, increases its acceptance.

[0086] The sole purpose of this description is to provide illustrative examples and to highlight further advantages and particular features of the present invention. Therefore, this description should not be interpreted as limiting the scope of application of the present invention or the patent rights claimed in the appended claims. In particular, features disclosed in conjunction with the method described herein can reasonably be used to extend the method described herein, and vice versa.

Claims

1. A computer-implemented method for identifying mechanical damage to a machine (20) based on performance indicators present at a rotational speed of the machine (20), the method comprising the steps of: - during a calibration period, receiving (S1) a plurality of calibration data points (kDP), said calibration data points respectively comprising a rotational speed of said machine (20) and a performance characteristic value assigned to said rotational speed, said calibration data points being determined based on sensor data of the machine (20) operating at one or more operating points; - calibrating (S) an AI-based Gaussian process regressor (GPR) by inputting calibration data points (kDP, 12) received during the calibration period, and outputting an estimated performance characteristic curve (gKPI, 11) and uncertainty ranges (13) assigned to each value of the performance characteristic curve (gKPI, 11) for all speed values ​​within a specified speed range indicative of a normal state of the machine (20); - receiving (S3) the data points (bDP) to be evaluated during an evaluation period following said calibration period; and - comparing (S4) the data point to be evaluated (bDP) with the estimated performance characteristic curve (gKPI, 11); and - outputting (S5) an exception message (AM) if the data point (bDP) to be evaluated lies outside at least one specified limit value of an uncertainty range (13) depending on the performance characteristic value.

2. The method according to claim 1, wherein The Gaussian process regressor (GPR) includes a statistical model for regression and prediction of data points, wherein the statistical model is a covariance function for modeling the relationship between input values ​​and output values, wherein hyperparameters of the covariance function are set by the calibration.

3. The method according to any one of the preceding claims, wherein The performance characteristic curve (gKPI, 11) is the mean value determined by the Gaussian process regressor (GPR), and the uncertainty range is the standard deviation determined by the Gaussian process regressor.

4. The method according to claim 3, wherein: The distance of the data point (bDP) to be evaluated from the mean value is calculated and compared with the standard deviation of the data point (bDP) on the estimated performance characteristic curve (gKPI, 11).

5. The method according to any one of the preceding claims, wherein The performance characteristic curve (gKPI, 11 ) includes performance characteristic values ​​for all speed values ​​within a specified speed range.

6. A method according to any one of the preceding claims, wherein: The Gaussian process regressor (GPR) is recalibrated with additional calibration data points (kDP) at specified time intervals.

7. A method according to any one of the preceding claims, wherein: During the calibration period, an increased number of calibration data points (kDP) are received per unit time relative to the evaluation period.

8. A method according to any one of the preceding claims, wherein The rotational speed is derived from magnetic field data, and the performance characteristic value is derived from vibration data measured at the machine (20).

9. The method according to any one of the preceding claims, wherein: If the number of abnormal messages (AM) for a data point (bDP) received within a specified time period exceeds a specified limit, a deterioration message is output.

10. The method according to any one of the preceding claims, wherein On a graphical user interface, the estimated performance characteristic curve (gKPI, 11) and the uncertainty range (13) assigned to each value of the performance characteristic curve (gKPI, 11) and the data point (bDP) to be evaluated are output.

11. The method according to any one of the preceding claims, wherein The mechanical damage is bearing damage or shaft misalignment of the machine (20) or imbalance of the machine (20).

12. A condition monitoring device (20) for identifying mechanical damage to a machine (20) based on performance indicators present at a rotational speed of the machine (20), the condition monitoring device comprising at least one processor configured to perform the following steps: - during a calibration period, receiving (S1) a plurality of calibration data points (kDP), said calibration data points respectively comprising a rotational speed of said machine (20) and a performance characteristic value assigned to said rotational speed, said calibration data points being determined based on sensor data of the machine (20) operating at one or more operating points; - calibrating (S) an AI-based Gaussian process regressor (GPR) by inputting calibration data points (kDP, 12) received during the calibration period, and outputting an estimated performance characteristic curve (gKPI, 11) and uncertainty ranges (13) assigned to each value of the performance characteristic curve (gKPI, 11) for all speed values ​​within a specified speed range indicative of a normal state of the machine (20); - receiving (S3) the data points (bDP) to be evaluated during an evaluation period following said calibration period; and - comparing (S4) the data point to be evaluated (bDP) with the estimated performance characteristic curve (gKPI, 11); and - outputting (S5) an exception message (AM) if the data point (bDP) to be evaluated lies outside at least one specified limit value of an uncertainty range (13) depending on the performance characteristic value.

13. A computer program product comprising a non-volatile computer-readable medium directly loadable into the memory of at least one digital computer, said non-volatile computer-readable medium comprising program code portions which, when executed by said at least one digital computer, cause said at least one digital computer to perform the steps of the method according to any one of claims 1 to 11.

Images