SYSTEM AND METHOD FOR CONDITION DETECTION OF A DEVICE

Description

[0001] In particular, the present disclosure relates to a system and method for detecting the condition of a device.

[0002] It is regularly necessary to automatically monitor the condition of devices. This typically involves recording measurement data that is meaningful for the condition of the device. This often requires searching for and analyzing specific signatures in the measurement data. For example, a change in the properties of the signature can indicate an impending failure of the device. For example, in a device in the form of a drive machine, a vibration that becomes more intense over several acceleration maneuvers can indicate an increasingly worn bearing and therefore a condition of the device that requires repair. Condition monitoring can therefore be used to initiate repairs in good time, before the bearing fails and, for example, blocks the drive machine.

[0003] However, detecting signatures in measurement data is often error-prone. As a result, the condition of the corresponding device is often not correctly identified. A major reason for errors in automated signature detection is often inaccurate measurement data, particularly those affected by measurement noise and / or general statistical measurement inaccuracies.

[0004] US 2008 / 015726 A1 relates to a fault diagnosis solution, whereby once a fault is identified by a fault detector, the fault is analyzed by a fault diagnostic device by comparing the fault with a collection of fault signatures stored in a fault diagnostic database. Each fault signature represents process conditions or process variables that are representative of specific faults.

[0005] US 2007 / 265713 A1 concerns the detection of an anomaly in data from components of an industrial process, whereby a predictive model is selected according to the detected operating mode of the respective component, and estimated data is generated using the predictive model. The estimated data and the operating data are compared, and an anomaly is detected if a specific anomaly criterion is met.

[0006] WO 2011 / 056144 A1 relates to a system for fault diagnosis in a power grid system, wherein a fault signature list is created from error messages of a monitoring system, from which a fault signature sequence (FSS) is formed through a serialization process. The fault signature sequence (FSS) is then divided into several subgroups to diagnose faults at the level of the entire system, the substation, or the circuit line.

[0007] However, in some applications, particularly in the aviation sector, it is regularly necessary to have the most precise automatic status detection of one or more devices in order to exclude an unexpected failure of one or more devices as far as possible.

[0008] The object of the present invention is to improve the detection of the state of a device by means of pattern recognition in faulty measurement data. According to one aspect, a system for automatically detecting the state of a device is provided. The system comprises a signal acquisition device for detecting faulty measurement data of a physical quantity describing the device. For example, the system comprises a sensor for detecting the physical quantity or receives measurement signals from a corresponding sensor of the device.The system further comprises an analysis unit for finding a predetermined pattern in the measurement data acquired by the signal acquisition device, wherein the analysis unit is configured to a) compare at least two different pattern sections of the predetermined pattern separately from one another with the measurement data, b) detect at least one position of each of the pattern sections in the measurement data based on the respective comparison, c) detect the predetermined pattern at one or more positions in the measurement data based on the detected positions and the order of the positions of the pattern sections, and d) determine the state of the device based on the one or more positions of the predetermined pattern.

[0009] This is based on the realization that the recognition of a given pattern in error-prone, particularly noisy, measurement data often leads to incorrect recognition of the pattern, i.e. to the erroneous recognition of the pattern at a position in the measurement data where the pattern is not actually present. To do this, the given pattern is divided into at least (or exactly) two sections, which are searched for independently of one another in the measurement data. If, after this independent search, the pattern sections are found in the correct order (corresponding to the order of the pattern sections in the given pattern) and, for example, adjacent to one another in the measurement data, it can be assumed with a particularly high degree of probability that the given pattern is actually present at the corresponding position in the measurement data. The independent search can significantly reduce the rate of false recognitions of the pattern.The state of the device can be determined based on the one or more positions of the predefined pattern in the measurement data, for example, by analyzing the measurement data at the one or more positions, for example, by comparing it with a threshold value, and / or by comparing the measurement data at several determined positions of the predefined pattern with each other. In a particularly simple example, the state of the device can be determined simply by recognizing the position of the predefined pattern in the measurement data itself, for example, if the predefined pattern represents a signature for a specific condition, such as a defect.

[0010] For example, the device is a battery. The physical quantity is alternatively or additionally an electrical quantity, in particular an electrical voltage. Batteries, in particular those that provide comparatively high currents to operate a drive device (for example a vehicle, e.g. an aircraft), undergo usage-dependent aging processes and can develop defects with increasing age or heavy use. Furthermore, production defects can also occur. If such a defect or error is not detected in time, the battery can fail and no longer provide energy. This can be counteracted with short replacement intervals, but this can be associated with corresponding downtimes and high expenditure. The system described here, on the other hand, can detect the condition of the battery particularly reliably.The predefined pattern describes, for example, the temporal progression of the electrical voltage of a battery cell of the battery (or of the entire battery) during a discharge process. By evaluating the measurement data at one or more detected positions of the predefined pattern, the condition of the device (e.g. the battery) can be determined particularly precisely, e.g. by detecting a systematic (in particular systematically increasing) deviation of the course of the measurement data at the position(s) from the predefined pattern. From this, it can be determined, for example, that the condition requires maintenance or replacement. In general, the condition can be determined as one of several predetermined states (e.g. one of two predetermined states, such as functional and defective; or one of three predetermined states, such as functional, requiring maintenance, and defective).

[0011] The analysis unit can be configured to compare the at least two different pattern sections of the specified pattern with the measured data using dynamic time warping (DTW). Dynamic time warping (DTW) involves mapping value sequences of different lengths onto one another. This allows, for example, differently stretched sequences of the measured data to be compared with the specified pattern. In the battery example, these can correspond, for example, to a slow and a fast discharge process.

[0012] Optionally, the analysis unit is configured to apply the dynamic time normalization to at least one of the at least two different pattern sections of the predefined pattern in order to generate one (or more) modified pattern sections. Thus, it may be sufficient to store only one predefined pattern for one type of characteristic profile of the measurement data (e.g., a battery discharge process).

[0013] Furthermore, the analysis unit can be configured to detect the at least one position of each of the pattern sections multiple times. In this case, at least one or at least two boundary conditions can be changed over the multiple implementations and a number of positions of the specified pattern found in the measurement data can be determined for each implementation. A number of positions found most frequently in the multiple implementations can be determined as the result for the most likely actually applicable number of positions of the specified pattern in the measurement data. Possible boundary conditions are, for example, the width of a search window in the measurement data for comparison with the pattern sections and / or a number of maximum findable positions of the pattern sections and / or the specified pattern.

[0014] Optionally, the device can be integrated into an aircraft. This is where the aforementioned advantages become particularly apparent.

[0015] At least one position of each of the sample sections in the measurement data can be determined by determining a minimum total deviation (e.g., a minimum Euclidean distance) between the respective sample section and the measurement data at multiple positions in the measurement data. This allows for fast and precise determination.

[0016] The measurement data, for example, is time series data. The one or more positions of the given pattern can each indicate a point in time or a period of time in the measurement data.

[0017] The system may further comprise a display device. The analysis unit is optionally configured to output the determined state of the device on the display device. Alternatively or additionally, the analysis unit is communicatively connected to the device. The analysis unit is configured, for example, to control and / or shut down the device based on the determined state of the device.

[0018] Optionally, the analysis unit comprises an artificial neural network and is configured to train the artificial neural network using the one or more positions of the predefined pattern in the measurement data. In this way, a particularly reliably trained neural network can be obtained, which allows a particularly accurate determination of the state of the device. The position detection of the predefined pattern allows for automatic labeling of the measurement points that define the pattern. Such labeling enables automatic preparation for the training of neural networks.

[0019] According to one aspect, a method for automatically detecting the state of a device is provided. The method can utilize the system according to any embodiment described herein. The method comprises acquiring, by means of a signal acquisition device, erroneous, e.g., noisy, measurement data of a physical quantity describing the device, and locating, by means of an analysis unit, a predetermined pattern in the measurement data acquired by the signal acquisition device.This finding in turn comprises comparing at least two different pattern sections of the predetermined pattern separately from one another with the measurement data, detecting at least one position of each of the pattern sections in the measurement data based on the respective comparison, detecting the predetermined pattern at one or more positions in the measurement data based on the detected positions and the order of the positions of the pattern sections, and determining the state of the device based on the one or more positions of the predetermined pattern.

[0020] Optionally, the method further comprises outputting, by means of the analysis unit, the determined state of the device on a display device.

[0021] Embodiments will now be described by way of example with reference to the figures, in which: Figure 1 shows an aircraft in the form of an airplane with a battery and a propeller electrically driven by the battery; Figure 2 shows a device and a system for automatically detecting the state of the device; Figures 3 and 4 show diagrams with various measurement data of an electrical voltage of the battery of the aircraft according to Fig. 1 ; Figures 5 and 6 Diagrams explaining the recognition of a pattern in measurement data; Figures 7 and 8 show the number of positions of a given pattern in the measurement data determined under different conditions according to Fig. 3 and Figure 4; and Figure 9 shows a method for automatically detecting the state of a device.

[0022] Figure 1 shows an aircraft 3 in the form of an electrically powered airplane. The aircraft 3 includes a propeller 31 driven by an electric machine 30 in the form of an electric motor.

[0023] The aircraft 3 further comprises a device in the form of a battery 2. The electric machine 30 is electrically connected to the battery 2 and is supplied with energy thereby. Optionally, the electric machine 30 serves as a generator and is configured to supply electrical energy to the battery 2.

[0024] Due to environmental influences, use, and / or age, the battery 2 may be in a deteriorated condition compared to its proper state. Furthermore, the battery may be in a critical condition, which is likely to be followed relatively quickly by a failure. To automatically monitor the condition of the battery 2, the aircraft 3 includes a system 1 for automatically detecting the condition of a device, specifically the battery 2 in the example shown.

[0025] Figure 2shows a schematic representation of the battery 2 and the system 1 for automatic condition detection.

[0026] The system 1 comprises a signal acquisition device 10 and an analysis unit 11. In Figure 2 These two components are shown separately, but they can of course also be implemented using common hardware and / or software.

[0027] The signal detection device 10 is operatively connected to the battery 2, in this case such that the signal detection device 10 can detect a voltage of the battery 2, in particular a voltage of a cell or, as in the example shown, the voltage of all battery cells (or a voltage of each cell) of the battery. This cell voltage indicates the charge state of the cell. When a certain power is delivered by the cell, the voltage drops with a characteristic curve. Deviations from this characteristic curve can indicate a deterioration in the condition of the cell.

[0028] In general, the signal acquisition device 10 is designed to acquire measurement data of a physical quantity (the cell voltage) of a device (the battery 2).

[0029] The signal acquisition device 10 provides the acquired measurement data to the analysis unit 11. For this purpose, the signal acquisition device 10 in the present case comprises an analog-to-digital converter. The analysis unit 11 is designed to find a predetermined pattern in the measurement data acquired by the signal acquisition device 10 by, as will be explained in more detail below, comparing at least two different pattern sections of the predetermined pattern separately from one another with the measurement data, detecting at least one position of each of the pattern sections in the measurement data based on the respective comparison, detecting the predetermined pattern at one or more positions in the measurement data based on the detected positions and the sequence of the positions of the pattern sections, and determining the state of the device based on the one or more positions of the predetermined pattern.

[0030] For this purpose, the analysis unit 11 in the example shown comprises a processor 110 and a memory 111 for storing instructions executable by the processor.

[0031] The system 1 further comprises an (optional) display device 12 for displaying the state determined by the analysis unit 11.

[0032] Optionally, the analysis unit 11 is communicatively connected to the device (here, the battery 2). The analysis unit 11 is designed to control, in particular to regulate, the device based on the determined state of the device (here, the battery 2). For example, the analysis unit 11 is designed to switch off the device based on the determined state of the device (here, the battery 2), e.g., to open an electrical circuit enclosing the device.

[0033] Figure 3shows measurement data D of the cell voltage of battery 2 over a period of a flight of aircraft 3. Full thrust was requested several times, specifically five times, over a period of time, which each time resulted in energy being drawn from battery 2. This is recognizable in the diagram shown by five characteristic signatures. The signatures each extend from a beginning B to an end E. The precise progression of these signatures, e.g., in comparison with each other and / or in comparison with a reference signature, allows conclusions to be drawn about the condition of battery 2, in particular whether the battery requires maintenance or is approaching the end of its service life.

[0034] Figure 4 shows measurement data D of the cell voltage of battery 2 over a period of another flight of aircraft 3. There, full thrust was requested twice.

[0035] Top right of the diagram of the Figure 4the given pattern M is shown, which is compared by the analysis unit 11 with the measurement data D in order to determine the positions of the pattern M in the measurement data D, in Figure 3 the five positions in Figure 4 the two positions. This also shows a division of the given pattern M into two halves, M1 and M2.

[0036] For example, a DTW algorithm is applied during the comparison to account for thrust requests of different lengths and / or requests for thrust less than full thrust and / or to account for different absolute values ​​of the measured data. For example, using DTW, the respective sample section M1, M2 (alternatively or additionally, the entire specified sample M) is scaled, for example, along the time axis and / or along the axis of the measured value at the respective time, to obtain a modified sample section M2'.

[0037] The search for the individual sample sections M1, M2 in the measurement data D is carried out below using the Figure 5 and 6 explained, whereby for illustration purposes very different measurement data D are shown, each in diagram (a) of the Figure 5 and 6 . As an example, the amplitude of a physical quantity is plotted against time.

[0038] The respective diagram (b) shows a sample section M3, M4 of a given pattern. In order to find this sample section M3, M4 in the measurement data D once or several times, the sample section M3, M4 is compared step by step with the measurement data D (with each section of the measurement data D). In this way, the total deviation of the sample section M3, M4 from a specific section of the measurement data D is determined, e.g., in the form of the sum of the absolute deviations, SAD, as shown in the respective diagram (c). The point at which the total deviation has a minimum corresponds to the most probable position of the sample section M3, M4 in the measurement data D. If multiple occurrences of the sample section M3, M4 are determined and / or permitted in the determination, the most probable positions are the corresponding number of the smallest minima. In the Figure 5 and 6 the respective beginning B of such a pattern section M3, M4 identified in the measurement data is marked.

[0039] The respective diagram (d) shows the measurement data D with the sample sections M3, M4 superimposed.

[0040] This search for one or more positions of the pattern sections M3, M4 is carried out by the analysis unit 11 for all (e.g., two) pattern sections M1-M4 of the measurement data D. In doing so, more occurrences of the separate pattern sections M1-M4 can sometimes be found in the measurement data D than actually exist. However, if a neighboring sequence of (all) pattern sections M1-M4 of the given pattern M is determined in the correct order, this is a particularly reliable indication of the presence of the given pattern. A character can be assigned to each pattern section M1-M4, so that the sequence of the pattern sections M1-M4 can be processed by the analysis unit 11 in the form of a character string. This allows for particularly efficient and fast processing.

[0041] With further reference to Figure 4 The sequence M1, M2 will therefore be found twice. This calculation can be performed automatically and quickly using processor 110 and, unlike the use of a neural network for this purpose, is comprehensible and reproducible. Furthermore, no training is required. The algorithm is therefore generally better suited for aviation certification than an algorithm based on neural networks. In contrast, training a neural network regularly requires a large number of "faulty" and "good" events.

[0042] The positions of the given pattern M in the measurement data D are indicated, for example, by the respective beginning B and / or the respective end E (in particular the corresponding time).

[0043] The search can be further refined by varying the search conditions. Figure 7 and 8 The respectively determined number of adjacent sequences of (all) pattern segments M1-M2 of the given pattern M in the correct order when searching with a varied number of permitted maximum minima and the window width of the measured data D and / or the pattern segment M1-M2 when compared with the respective pattern segment M1-M2. However, these are merely examples of possible framework conditions.

[0044] In both cases, a plateau is found at a number of five ( Figure 7 ) or two ( Figure 8) detected occurrences of the given pattern M in the measurement data D. The number indicated by the plateau can be used as the result for the number of detected occurrences of the given pattern M in the measurement data D. This enables a particularly robust determination of this number. In this way, false positive patterns can be largely or completely excluded. This is especially true if the acquired measurement data D is noisy, e.g., due to environmental influences or electromagnetic interference from surrounding components, as these can cause false minima.

[0045] Since the number can be determined robustly in this way, it is possible to determine the corresponding positions in the measurement data with particular precision. Thus, the analysis unit 11 can be configured to perform the search several times under varying conditions and then search for a plateau and / or a most frequently occurring value in the obtained results of the number of positions.

[0046] Figure 9 shows a method for automatic state detection of a device (e.g. the battery 2), comprising: Step S1: Acquiring, by means of the signal acquisition device 10, erroneous measurement data D of a physical quantity describing the device (battery 2) (e.g., cell voltage). Step S2: Locating, by means of the analysis unit 11, the predetermined pattern M in the measurement data D acquired by the signal acquisition device 10. Step S2 comprises the sub-steps S21 to S24. Step S21: Comparing at least two different pattern sections M1-M4 of the predetermined pattern M separately from one another with the measurement data D. Step S22: Acquiring at least one position of each of the pattern sections M1-M4 in the measurement data D based on the respective comparison. Step S23: Recognising the predetermined pattern M at one or more positions in the measurement data D based on the acquired positions and the order of the positions of the pattern sections M1-M4.Step S24: Determining the state of the device (battery 2) based on the one or more positions of the predetermined pattern M. Step S3: Outputting, by means of the analysis unit 11, the determined state of the device (battery 2) on a display device 12 and / or controlling the device (battery 2) based on the determined state.

[0047] Above, particular attention was paid to a cell voltage of a battery 2, wherein, for example, a measurement of other electrical or mechanical quantities, e.g. phase currents, a (different) DC or AC voltage, a current intensity, a speed and / or a torque as a respective physical quantity is also possible, e.g. of electric aircraft propulsion systems, and / or a derived quantity such as mechanical power, electrical power, active power, apparent power, cosine phi, efficiency and / or a controlled variable of the respective device. In the case of electric aircraft propulsion systems, the following devices are particularly conceivable: motor, inverter, battery, energy / power distribution device and / or controller. It is understood that the invention is not limited to the embodiments described above and various modifications and improvements can be made without deviating from the concepts described here. List of reference symbols

[0048] 1System 10Signal acquisition device 11Analysis unit 110Processor 111Memory 12Display device 2Battery 3Aircraft 30Electric machine 31Propeller BBegin DMeasurement data EEnd Mspecified pattern M2'modified pattern section M1-M4Pattern section

Claims

1. System (1) for automatic state detection of a device (2), comprising: - a signal detection device (10) for detecting erroneous measurement data (D) of a physical quantity describing the device (2); and - an analysis unit (11) for finding a predetermined pattern (M) in the measurement data (D) acquired by the signal detection device (10), wherein the analysis unit (11) is designed to ∘ compare at least two different pattern sections (M1-M4) of the predetermined pattern (M) separately with the measurement data (D), ∘ based on the respective comparison, record at least one position of each of the pattern sections (M1-M4) in the measurement data (D), ∘ based on the recorded positions and the sequence of the positions of the pattern sections (M1-M4), recognize the predetermined pattern (M) at one or more positions in the measurement data (D) and ∘ determine the state of the device (2) based on the one or more positions of the predetermined pattern (M).

2. System (1) according to Claim 1, wherein the device (2) is a battery and the physical quantity is an electrical voltage.

3. System (1) according to Claim 1 or 2, wherein the analysis unit (11) is designed to compare the at least two different pattern sections (M1-M4) of the predetermined pattern (M) with the measurement data (D) by means of dynamic time warping.

4. System (1) according to Claim 3, wherein the analysis unit (11) is designed to apply the dynamic time warping to at least one of the at least two different pattern sections (M1-M4) of the predetermined pattern (M) in order to generate at least one modified pattern section (M2').

5. System (1) according to one of the preceding claims, wherein the analysis unit (11) is further designed to perform the detection of the at least one position of each of the pattern sections (M1-M4) several times, wherein at least two framework conditions are changed over the multiple executions and a number of found positions of the predetermined pattern (M) in the measurement data (D) is determined for each execution, wherein a number of found positions determined most frequently in the multiple executions is determined as the result for the number of positions of the predetermined pattern (M) in the measurement data (D).

6. System (1) according to one of the preceding claims, wherein the device (2) is part of an aircraft (3).

7. System (1) according to one of the preceding claims, wherein the at least one position of each of the pattern sections (M1-M4) in the measurement data (D) is determined by determining a minimum total deviation between the respective pattern section (M1-M4) and the measurement data (D) at a plurality of positions of the measurement data (D).

8. System (1) according to one of the preceding claims, wherein the measurement data (D) are time series data and the one or more positions of the predetermined pattern (M) each indicate a point in time or time period in the measurement data (D).

9. System (1) according to one of the preceding claims, further comprising a display device (12), wherein the analysis unit (11) is designed to output the determined state of the device (2) on the display device (12).

10. System (1) according to one of the preceding claims, wherein the analysis unit (11) comprises an artificial neural network and is designed to train the artificial neural network by means of the one or more positions of the predetermined pattern (M) in the measurement data (D).

11. Method for automatic state detection of a device (2), comprising: - detecting (S1), by means of a signal detection device (10), erroneous measurement data (D) of a physical quantity describing the device (2); - finding (S2), by means of an analysis unit (11), a predetermined pattern (M) in the measurement data (D) acquired by the signal detection device (10), comprising: ∘ comparing (S21) at least two different pattern sections (M1-M4) of the predetermined pattern (M) separately with the measurement data (D), ∘ detecting (S22) at least one position of each of the pattern sections (M1-M4) in the measurement data (D) based on the respective comparison, ∘ detecting (S23) the predetermined pattern (M) at one or more positions in the measurement data (D) based on the detected positions and the order of the positions of the pattern sections (M1-M4) and ∘ determining (S24) the state of the device (2) based on the one or more positions of the predetermined pattern (M).

12. Method according to Claim 11, further comprising: outputting (S3), by means of the analysis unit (11), the determined state of the device (2) on a display device (12).