Computer-implemented method for controlling system based on detected trend

Abstract

To provide a computer-implemented method for controlling a system based on a trend detected in a data time series acquired by a sensor.SOLUTION: A method includes the steps of: detecting and providing a data time series by a sensor (S10); applying at least one linear filter function with a resolution L to each point in time t of the data time series (S12); obtaining a characteristic value for quantifying the temporal development of the data time series for each resolution L (S14); detecting a trend in the data time series when one of the characteristic values of the data time series reaches a threshold value (S16); and triggering a response of the system when a trend has been detected in the data time series (S18).SELECTED DRAWING: Figure 4

Description

[Technical field]

[0001] The present invention relates to the field of automated trend identification in data and control of systems based on identified trends. [Background technology]

[0002] Identifying trends in a system's workflow can be important, for example, to identify wear, abnormal behavior or the need for maintenance of the system. In particular, the earlier a trend that is detrimental to the system is identified, the better one can control against the trend or take steps to eliminate the trend. It is becoming increasingly important to automatically extract such trends from data that becomes available early.

[0003] To this end, various approaches exist for automatically identifying trends in data time series.

[0004] Linear trend estimation is known from Bianchi et al., “A comparison of methods for trend estimation”, Applied Economics Letters 6.2 (1999): 103-109. The method presented there is used in statistics to decompose a data time series into trend, seasonality and noise. There are two types of trend identification in the method: model-based trend identification and non-parametric trend identification. In the non-parametric trend identification, for example, a convolution between a triangular filter and a finite symmetric window of the data time series is applied.

[0005] From Percival et al., “Wavelet methods for time series analysis”, Vol. 4, Cambridge University Press, 2000, wavelet analysis is known, which is compared to Fourier analysis. Here, for this purpose, wavelets are used, which represent local oscillations around zero, instead of sine and cosine functions. The signal and the wavelets are then convolved with each other, each with a different resolution.

[0006] Prewitt, Judith, "Object enhancement and extraction", Picture processing and Psychopictorics 10.1. 1970: 15-19, discloses a filter for finding edges in image processing. The so-called Prewitt operator consists of two 3x3 kernels that are convolved with a grayscale image.

[0007] Another approach is to use deep learning, which has the drawback of being relatively resource-intensive to apply: the computational operations are complex, which not only increases the power consumption of the system on which it is implemented, but also occupies computational capacity that could be used for other methods, or generally requires a larger system. [Prior art documents] [Non-patent literature]

[0008] [Non-Patent Document 1] Bianchi et al., ““A comparison of methods for trend estimation,” Applied Economics Letters 6.2 (1999): 103-109.” [Non-Patent Document 2] Percival et al., “Wavelet methods for time series analysis”, Vol. 4, Cambridge University Press, 2000. [Non-Patent Document 3] Prewitt, Judith, ““Object enhancement and extraction”, Picture processing and Psychopictorics 10.1. 1970: 15-19” Summary of the Invention [Problem to be solved by the invention]

[0009] The problem underlying the present invention is therefore to propose a resource-friendly method for trend identification and the associated control of the system. [Means for solving the problem]

[0010] The above problem is solved by the subject matter of the respective independent claims.

[0011] Disclosure of the Invention According to a first aspect of the present invention, the above problem is solved by a computer implemented method for controlling a system based on trends detected in a data time series detected by a sensor, the method comprising the steps of: - detecting and providing a data time series by a sensor; - applying at least one linear filter function with a resolution L at each time instant t of the data time series and obtaining, for each resolution L, a characteristic value quantifying the temporal evolution of the data time series; - detecting trends in the data time series when one of the characteristic values ​​of the data time series reaches a threshold value; - triggering a system reaction when a trend in the data time series is detected; The problem is solved by a computer-implemented method, comprising:

[0012] The detected data time series is in particular a time series of one-dimensional data, such as a temperature progression, a filling level progression, a pressure progression or similar quantities, in other words the data time series comprises the values ​​of a measurable quantity detected in a particular time section.

[0013] The quantities are detected by sensors to detect trends in the workflow. Thus, the sensors may be any commercially available sensor that detects a measured quantity, such as, but not limited to, a temperature gauge, a pressure sensor, a fill level gauge, or an electronic measuring device, such as an ammeter, voltage meter, or resistance meter. The sensors may be part of the system or may monitor parameters external to the system.

[0014] A linear filter function is applied to the data time series for each time point. The application of the filter function to the data time series includes, inter alia, mathematical operations on the data time series. The application of the filter results in a new function that can be analyzed. Such an analysis allows the application of the filter to the data time series to be quantified. In particular, the average slope of the new function gives an indication of the presence of a trend. As a result, from the new function, individual characteristic values ​​for each resolution are calculated that are suitable for quantifying the temporal evolution of the data time series.

[0015] In this case, the characteristic value can preferably be compared with a threshold value, which can preferably be selected such that small trends in the data time series are not identified as trends within the meaning of the present invention. This makes it possible to prevent, for example, small trends that are within the tolerance range of the system from not triggering a prepared action. In other words, the threshold value can be selected such that the typical measurement error of the sensor, the play of any movable parts possibly present in the system, or the tolerances of the system in general are taken into account.

[0016] When the characteristic value exceeds the threshold value, this necessarily leads to triggering the reaction of the system.Because the threshold value already takes into account the error range in the detection of the data time series and the normal operation of the system, the actual trend that requires the system to respond occurs in the data time series.For example, the temperature may rise or fall slowly after a certain fluctuation, which allows the system to react accordingly, for example, by inputting more or less power.

[0017] In particular, a fixed threshold value can be used if the error range is known based on physical system knowledge. If the values ​​in the abnormal range are outside of the normal or expected working range, the threshold value can be determined statistically, possibly from past observations. Alternatively, the threshold value can be calibrated in a validation data set, which ensures that known anomalies are detected with the smallest possible error alarm rate (false positive rate).

[0018] In this case, the response of the system may depend on the system itself and / or the data time series being monitored. A trend does not necessarily have a negative effect on the system. It may have a positive effect. For example, if a new machine is in operation and machine vibration is detected in the data time series, the vibration will decrease over time as the machine starts to operate, for example, until lubricant is fully distributed among the parts during normal operation.

[0019] Since the application of a linear filter function to a data time series places very low demands on system resources, the method according to the present invention is system independent and suitable for identifying trends in a data time series that trigger a system response in a resource-friendly manner.

[0020] In one embodiment, the trend to be identified is an upward or downward slope.

[0021] Depending on whether an upward or downward trend is identified, filters with opposite signs can be implemented, which makes the realization of the corresponding filters particularly simple.

[0022] The detection of trends is direction-dependent. The expected trend direction also depends on the system being monitored and the parameters being measured. For example, if the state of a battery cell is monitored, then essentially several parameters quantifying the state may be appropriate. Temperature or output power are the parameters here. Over time, the temperature at the time of power discharge may for example increase. That is, during the operation of the battery, a trend of increasing temperature over time is discernible. Conversely, the battery output may decrease as time progresses. If the output is observed over a long period of time, a downward trend in the battery output is observable. However, in the particular case, the charging cycle of the battery should be taken into account, since a fully charged battery usually has a higher power discharge than a half-empty battery.

[0023] To do this properly, a quantification of the data time series is compared to a threshold, which causes fluctuations on small time scales to be largely ignored.

[0024] In one embodiment, the step of applying the filter function with respect to the resolution L comprises the following steps: - setting a resolution L; - a linear function phi with L values L (t); - determining a section of the data time series for a range tL to t having L values; -Linear function phi L multiplying (t) by a section of the data time series; For all L values, the linear function phi L integrating all products of (t) and the data time series; Includes.

[0025] The data time series is decomposed into sections, each containing L sections, for multiplication with the filter function, such that each section f(t) has L values ​​from f(t-L+0) to f(t-L+L)=f(t).

[0026] The resolution L determines how many values ​​of the data series are examined for a trend. Depending on the system and the process being monitored, the resolution L can be chosen in various ways. The sensors used and the data series being examined can also affect the resolution. For example, if measurements are taken once a day and the data series is formed from values ​​over a period of weeks, months or years, a different, particularly lower, resolution may be appropriate. In contrast, if vibrations in a motor of a system are monitored at the millisecond or microsecond level, the examined section of the data series may contain a number of values ​​several times larger. That is, a higher resolution is desirable, so that more values ​​can be considered in the search for a trend.

[0027] The linear filter is specifically constructed to increase or decrease in L increments from -1 to +1 depending on whether the trend to be identified is upward or downward. The linear function phi L (t) is multiplied by a section of the data time series f(t) and the resulting products are accumulated. That is,

number

[0028] The characteristic value can be detected in particular as an absolute value in order to also identify a downward trend. In this case, the characteristic value is K=|k|. On the other hand, a negative trend can also be indicated by a negative product k. The characteristic value is therefore evaluated not only for its absolute value but also for its sign. If K is below or above a threshold value S, a trend is detected.

[0029] In one embodiment, the individual products phi L (l)*f(l) are weighted with different strengths in the multiplication. For this purpose, the characteristic value K is calculated, for example, by the modified formula

number

number

[0030] The weighting of certain regions of the data series for trend detection advantageously serves to make linear trends easily detectable, in particular by monitoring edge regions of the examined data series, since these regions are particularly strongly interdependent when trends are in fact present.

[0031] In one embodiment, the system is a test bench for automobiles, and the system's reaction to detecting a trend includes shutting down the test bench. The data series may be, in particular, parameters to be examined of the vehicle being tested on the test bench, such as, but not limited to, exhaust gas values ​​or exhaust gas temperatures, and battery temperatures or battery power for electric vehicles.

[0032] When a trend is identified, it can indicate a functional error or a need for maintenance of the vehicle. The test bench itself is no longer necessary because the question of the need for maintenance is clear. Thus, the shutting down of the system saves energy and resources, since the vehicle currently being inspected can be brought to the workshop while another vehicle can possibly be put on the test bench. Furthermore, the vehicle currently being tested can be shut down as well, which will result in less exhaust gas generation or unnecessary strain on the battery.

[0033] In another embodiment, the sensor is a sensor that measures temperature, in which the data series may be the temperature of a subsystem of the vehicle, in particular the battery, fuel cell or drive unit.

[0034] Excessively high temperatures in vehicle subsystems usually indicate a threatening failure in the vehicle. If the battery is too hot, individual battery cells may age more quickly. If the exhaust gases are too hot, this indicates mis-combustion of fuel.

[0035] In another embodiment, the system is a vehicle and the sensor detects a parameter of the vehicle or a subsystem of the vehicle.

[0036] Advantageously, the proposed method is particularly suitable for execution by a vehicle's on-board computer, since the on-board computer is often equipped with very limited resources, and the resource-friendly detection of trends in data time series according to the invention is particularly suitable for on-board vehicle applications.

[0037] In one embodiment, the system is a movable member within a machine, in which case the sensor may be, inter alia, a sensor that detects vibrations of the member of the machine, in which case the reaction of the system may include shutting down the machine or identifying the machine as needing maintenance.

[0038] Just like all machines, maintenance must always be performed. For this, it is often difficult to estimate the actual wear, and maintenance may be performed that is not yet necessary. Maintenance is costly and time-consuming for the machine operator, so it is postponed when possible. An indicator of the need for maintenance allows maintenance to be performed at the most efficient time.

[0039] Machines with moving parts usually vibrate. In this case, the vibrations can be a fingerprint for the working process of the machine. When wear occurs, more friction occurs in the moving parts, for example due to lack of lubricant, wear of the lubricating plate or increased play between two parts. These effects make the vibrations in the machine stronger.

[0040] If the amplitude of vibrations of the machine is detected as a data series, the proposed method can be used to check the data series for trends. In this case, an upward trend in the data series is synonymous with stronger vibrations and thus greater wear. A torque indicating no longer acceptable wear can be set by means of a threshold value, which triggers a reaction of the system. In some cases, the machine can be shut off for safety reasons. In other cases, it is sufficient to mark the machine with an identifier. The machine can be specifically marked as "maintenance required" to inform the operator that the machine should in fact be maintained. Such an embodiment allows maintenance to be planned at the required time during the operation of the machine. This avoids the need to incur costs for maintenance that is not yet necessary, which makes the operation of the machine more favorable.

[0041] In other cases, the machine may be marked with an identifier "ready to operate". When a new machine or a machine that has just been maintained is put into service, it may first have to be set to ready to operate. For example, lubricant must first be distributed in the moving parts or a certain component must reach a target temperature. For example, a sensor may detect the vibration of the machine as a data time series. If the amplitude of the vibration decreases over time, this can be identified as a trend as long as the characteristic value of the vibration time series is above a corresponding threshold value. Such a downward trend may trigger, for example, an identifier of the machine "warming up", "getting ready", etc. However, if the characteristic value reaches a level below the threshold value, the identifier is erased and the machine is indirectly or directly marked with the identifier "ready to operate". The indirect identifier "ready to operate" may include, among other things, the cancellation of the identifier "not ready to operate" or an equivalent identifier.

[0042] In another embodiment, the system is a fuel cell, and the sensor detects the pressure in the fuel cell, the current delivered by the fuel cell, the temperature at or within the fuel cell, or generally any virtually calculated operating condition of the fuel cell, and in such an embodiment, the reaction of the system may include, among other things, reducing the power output of the fuel cell or shutting down the fuel cell.

[0043] Fuel cells are based on the principle of electrochemical reaction between a fuel and an oxidant, thus generating electrical energy. A fuel cell consists of several main components, including an anode, a cathode, and an electrolyte. The anode serves as the site of oxidation of the fuel, while the cathode is responsible for the reduction of the oxidant. The electrolyte allows the transport of ions between the anode and the cathode. The electrolyte layer between the anode and the cathode allows the flow of ions, which generates an electric current.

[0044] To ensure efficient and long-term power output of a fuel cell, it is crucial to select an appropriate catalyst. The catalyst should have high activity to promote the reactions at the anode and cathode. In addition, it is desirable for the catalyst to be stable and have good durability against the operating conditions of the fuel cell, including temperature, pressure, and fuel composition.

[0045] Each of the above-mentioned parameters can be detected by a corresponding sensor as a data time series, and if a trend is identified in one of these data time series, this may indicate abnormal and possibly inefficient operation of the fuel cell.

[0046] By shutting off or reducing the output of the fuel cell, the fuel cell can be protected against failure, or at least protected before failure occurs, depending on the situation. Additionally, the fuel cell can be specifically identified as "in need of maintenance."

[0047] In a further aspect, the invention relates to a computer program comprising program code for implementing, when the computer program is run on a computer, a method for controlling a system based on trends detected in a data time series detected by a sensor as described above.

[0048] In a further aspect, the invention relates to a computer readable data carrier comprising a program code of a computer program for performing, when the computer program is run on a computer, a method for controlling a system based on trends detected in a data time series detected by a sensor as described above.

[0049] In a further aspect, the present invention relates to a subsystem for a system performing a method for controlling the system based on a trend detected in a data time series detected by a sensor, the subsystem having means for performing the method for controlling the system based on a detected trend as described above.

[0050] The subsystem may be specifically configured as part of a system, for example the subsystem may be formed as a part of a fuel cell responsible for monitoring the detected data time series.

[0051] Generally, therefore, there is provided a computer-implemented method for controlling a system based on trends detected in a time series of data detected by a sensor, and an apparatus for performing the method.

[0052] The described configurations and developments can be combined with one another in any desired manner.

[0053] Other possible configurations, developments and realizations of the invention also include combinations not explicitly mentioned of the individual features of the invention described above or below with reference to the exemplary embodiments.

[0054] The accompanying drawings are provided for a further understanding of the embodiments of the present invention, the drawings illustrating the embodiments and used in conjunction with the explanation of the basic methods and concepts of the present invention.

[0055] Other embodiments and many of the above-mentioned advantages will become apparent with reference to the drawings, in which: Elements illustrated in the drawings are not necessarily drawn to scale relative to each other. [Brief description of the drawings]

[0056] [Figure 1] FIG. 2 illustrates a schematic flow of a method according to an embodiment. [Diagram 2] FIG. 2 is a schematic diagram illustrating a flow of a portion of a method according to an embodiment. [Figure 3a] FIG. 13 is a tabular diagram illustrating two example data series and their products with linear filters. [Figure 3b] FIG. 13 is a tabular diagram illustrating two example data series and their products with linear filters. [Figure 4] FIG. 1 illustrates, in graphical form, a data time series in which one or more trends have been identified. [Diagram 5] FIG. 2 illustrates a subsystem for performing a method according to one embodiment. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0057] In the various figures of the drawings, elements, modules or components having the same or similar functions are provided with the same reference numerals unless otherwise stated.

[0058] A flow diagram of a method according to an embodiment of the present invention is shown in Figure 1. The method starts at step S10 with the provision of a data time series detected by a sensor.

[0059] According to one embodiment, the data time series represents the time evolution of a parameter characterizing the system controlled within the framework of the method. The sensor may preferably be part of the system and transfers the data time series to a subsystem of the system being monitored here. This may allow the data time series to be examined for trends during the operation of the system.

[0060] In such an embodiment, the sensor may detect a parameter of a system different from the system for which the reaction is triggered as a data time series. For example, the sensor may detect a parameter of the vehicle, in particular the battery condition, and transmit this to a test bench, where a reaction is triggered when a trend is detected in the data time series of the vehicle parameter.

[0061] To carry out the method, the data time series may be temporarily stored. Preferably, the system has a dedicated main memory or an area in the main memory for this purpose. The amount of the main memory or the amount of main memory held determines the set of data points of the data time series that can be used for trend identification.

[0062] A linear filter function is applied to the detected data time series in step S12 for at least one resolution L. By applying the filter function to the data time series, a new quantifiable function results, from which a characteristic value is then obtained in step S14, on the basis of which an upward or downward trend can be identified. The characteristic value may in particular be the slope of a regression line through the new function.

[0063] In step S16, the characteristic value of the data series is compared with a threshold value to check whether it has been reached. In this case, the threshold value is selectively exceeded below or above depending on the system being monitored, the detected data series and the expected trend. If it is important to protect the system against an excessively strong increase in a parameter, for example an excessively strong increase in the vibration amplitude detected by a motion sensor, a trend is identified when the characteristic value exceeds the threshold value. However, if, for example, the output of the system is measured and the system is monitored so that it does not fall below a certain value, a downward trend is identified when the characteristic value falls below the threshold value.

[0064] If the characteristic value reaches the threshold value, a reaction of the system to the detection of a trend in the data time series is triggered in step S18. The reaction triggered depends on the system and the parameter being monitored. The reaction here may include specifically marking a system having a certain operating state as "maintenance required" or shutting down the system. In this embodiment, special processes can also be triggered if a trend is identified. For example, an additional drain valve of an intermediate reservoir for liquid can be opened if the filling level shows an upward trend.

[0065] If no trend is detected in comparing the characteristic values ​​to the thresholds, the method returns to step S10 and the process begins anew.

[0066] In this embodiment, a second round of the method with a second data time series can be performed, where some of the values ​​match the values ​​of the first data time series from the first round. This allows, for example, to continuously monitor a parameter and check for trends. For example, the method can be performed with a data time series containing 100 values. Since the first 100 values ​​are checked for trends in the first round and 50 new values ​​are found in the second round, the 100 values ​​of the data time series for the second round are composed of the 50 values ​​of the first data time series from the first round and the 50 newly found values. This allows trends to be detected with twice the time resolution.

[0067] In Fig. 2, the sub-steps of step S12 of Fig. 1 of applying a filter function are shown in a schematic manner according to one embodiment. In a first sub-step S12.1, a resolution is set for the data time series to be examined. The setting can be made individually, i.e. adjusted to the conditions of the monitored system and / or the external environment. In particular, one or more resolutions can be predefined. Alternatively, the resolution can also be systematically set the same for each round.

[0068] Next, we use the linear function phi to form a linear filter. L (t) is generated (S12.2). The function phi L (t) contains as many values ​​as the data time series or section of the data time series in which a trend is to be detected. The number of values ​​corresponds to the resolution L.

[0069] A data series is not fundamentally limited to a specific number of values. In the framework of the present invention, the data series is formed from a defined number L of values ​​from an infinitely detectable data series. This allows a trend to be checked for data series that are detected continuously at regular intervals. For this purpose, the values ​​of the data series can be continuously written into an intermediate memory, the oldest values ​​of the data series being overwritten by the newest values. Each time a trend in the data series is detected, the intermediate memory is read out. The data series of length L detected in this way is filtered in step S12.3 with a linear filter function phi L It is multiplied by (t).

[0070] In the final step S12.4 shown in FIG. 2, the linear filter function phi L The product of (t) and the data time series f(t) is integrated, resulting in a single value that can be used as a characteristic value in the subsequent method flow (see Figure 1).

[0071] A linear function phi with L values L In cases where (t) appears to rise linearly from -1 to +1, if no trend can be discerned in the data series, a characteristic value of 0 or close to 0 will result, since the individual products compensate each other. However, if a trend does exist, the characteristic value should be much greater or less than 0, since in this case later values ​​in the data series affect the characteristic value more than earlier values.

[0072] In Fig. 3a and 3b, the linear filter function phi L Two data series f(t) are shown in tabular form that are multiplied with (t) to form a new function u(t). Values ​​of the new function u(t) can be accumulated to form a characteristic value for the data series f(t).

[0073] In the data series f(t) shown in Figure 3a, an upward trend is easily discernible. This is especially visible from the graphical representation of the data series f(t). The data series f(t) in Figure 3b fluctuates back and forth between values ​​1 and 2, and therefore does not show any downward or upward trend.

[0074] In this example, the data series f(t) consists of five values, so the linear filter function phi L (t) is also divided into five steps. The filter function phi L (t) progresses linearly from -1 to +1 in increments of 0.5.

[0075] The data time series f(t) is filtered by a filter function phi L (t), we obtain a new function u(t), which is therefore u(t)=f(t)*phi L (t) It is calculated by:

[0076] Then, the characteristic value k of the data time series f(t) is calculated as the sum of the values ​​of the function u(t) as follows:

number

[0077] The characteristic value k for the data time series f(t) in Figure 3a is 2.5 (=-1+-1,5+0+2+3), whereas the characteristic value k for the data time series f(t) in Figure 3b is 0 (=-1+-1+0+1+1).

[0078] The feature value is compared to a threshold, which may be, for example, 1. Thus, all feature values ​​k above 1 or below -1 represent a trend in the data time series f(t). Furthermore, positive feature values ​​will represent an upward gradient and negative feature values ​​will represent a downward gradient.

[0079] However, it should be pointed out that the example data series f(t) shown in Figures 3a and 3b is greatly simplified for the sole purpose of demonstrating the basic principles of the invention. In order to be able to make statistically significant statements regarding trends, it is advantageous to use a data series with as many values ​​as possible.

[0080] Another data time series f(t) is shown with respect to time t in Figure 4. The data time series itself is shown by a dotted line. The data time series may be recorded during analysis or monitoring of a system and may include one or more processes.

[0081] In this example, the data time series f(t) was examined for an upward trend. An upward trend was detected in sections A, B, C, D, and E. It should further be noted that the upward trend in section C is steeper than the upward trend in section B. This can be identified, for example, by the characteristic value for section C being greater than the characteristic value for section B.

[0082] The data time series f(t) declines between sections A and B, between sections C and D, and between sections D and E. Although a downward trend can essentially be identified here, such a downward trend can be ignored in this embodiment by selecting positive characteristic values ​​above a threshold.

[0083] It should be noted that in addition to the upward trends in sections A to E, a long-term upward trend can also be detected over the entire data series f(t), summarized in section F. The line reflecting the trend is likewise plotted on the graph as a dashed line. This suggests that the trend does not have to be examined in only one time scale for the data series f(t). The data series f(t) can be monitored on different time scales in order to detect micro- or macro-trends. For this purpose, the individual values ​​of the data series can be stored in an intermediate memory for a relatively long period of time. If micro-trends are of interest, the trend can be examined for a portion of the data series. In contrast, if a macro-trend is to be detected, the trend is examined over a relatively long period of time and a correspondingly larger number of values ​​are read out of the system.

[0084] FIG. 5 illustrates, in a schematic and exemplary manner, an embodiment of a subsystem 10 for controlling a system based on trends detected in a data time series detected by a sensor 12 .

[0085] The subsystem 10 is configured to receive and process data time series from the sensor 12. For this purpose, the subsystem 10 is communicatively connected to the sensor 12. The communication connection can be wired or wireless. In particular, the sensor 12 or another subsystem containing the sensor 12 transmits the data time series, either as a continuous data stream or as packets containing multiple values, to an intermediate memory 14 of the subsystem 10. In the intermediate memory 14, the values ​​of the data time series are temporarily stored until they can be read out and further processed. The temporary storage can take place, for example, for a period of time during which sufficient data is available to perform trend detection.

[0086] The computation unit 16 can, according to the invention, fully or partially read the data time series from the intermediate memory 14 in order to detect trends in the data time series or parts thereof. The computation unit 16 can, for example, comprise a CPU, a GPU or any unit configured to perform computational operations.

[0087] The results of the check, i.e. information on whether or not a trend was detected and, if so, what trend and when, can be stored by the calculation unit 16 in another intermediate memory 18, which in some embodiments may be part of the intermediate memory 14. The two intermediate memories 14 and 18 can be configured as virtual intermediate memories of the subsystem 10, which can be located in a common hardware module or in different hardware modules separate from each other.

[0088] If the computing unit 16 detects a trend, the subsystem 10 sends a signal via the interface 20 to the system, which can then respond to the detected trend according to the signal. The interface 20 can be, for example, a USB interface or a wireless interface, in particular a Bluetooth interface or a WLAN interface. The signal can, for example, contain specific processing instructions for the system or activate processing instructions stored in the system, so that the system itself decides what reaction to trigger by the information about the detected trend. Especially the latter case is advantageous, since the subsystem can be combined with any system as an independent subsystem, if a suitable interface to the interface 20 is present in the system.

Claims

1. 1. A computer-implemented method for controlling a system based on trends detected in a data time series detected by a sensor, the method comprising the steps of: - detecting and providing a data time series by a sensor (S10); - applying (S12) at each instant t of said data series at least one linear filter function having a resolution L and obtaining (S14) for each resolution L characteristic values ​​quantifying the temporal evolution of said data series, - detecting a trend in said data time series when one of the characteristic values ​​of said data time series reaches a threshold value (S16); - triggering a reaction of the system when a trend in the data time series is detected (S18); A computer-implemented method comprising:

2. The computer-implemented method of claim 1 , wherein the trend is an upward gradient or a downward gradient.

3. The step of applying (S12) said filter function with respect to resolution L comprises the following steps: - a step of setting a resolution L (S12.1); - a linear function phi with L values Ｌ (t) (S12.2); - calculating a section of the data time series for a region t-L to t having L values; - the linear function phi Ｌ multiplying (t) by a section of the data time series (S12.3); - For all L values, the linear function phi Ｌ (S12.4) integrating all products of (t) and the data time series; 3. A computer-implemented method according to claim 1 or 2, comprising:

4. the system is a test bench for an automotive vehicle; The reaction of the system includes shutting down the test bench and / or the vehicle. A computer-implemented method according to any one of claims 1 to 3.

5. the sensor is a sensor for measuring temperature, the data time series determining a temperature of a subsystem of the vehicle; 5. The computer-implemented method of claim 4.

6. the system being a movable member within a machine; the sensor detects vibration of the member, The system's response may include shutting down the machine or marking the machine with an identifier, particularly if maintenance is required. A computer-implemented method according to any one of claims 1 to 5.

7. the system is a fuel cell; the sensor is a sensor for detecting a pressure in the fuel cell, a sensor for detecting a current output from the fuel cell, a temperature sensor in the fuel cell, or a virtual sensor for detecting an operating state of the fuel cell; the system's response includes reducing the output of the fuel cell or shutting off the fuel cell; A computer-implemented method according to any preceding claim.

8. 8. A computer program comprising a program code for implementing a method for controlling a system based on trends detected in a data time series detected by a sensor according to any one of claims 1 to 7, when the computer program is run on a computer.

9. 8. A computer readable data carrier comprising a program code for a computer program for implementing a method for controlling a system based on trends detected in a data time series detected by a sensor according to any one of claims 1 to 7, when the computer program is run on a computer.

10. A subsystem 10 for a system for implementing a method for controlling a system based on trends detected in a data time series detected by a sensor 12, comprising: The subsystem 10 comprises means for implementing a method for controlling the system on the basis of detected trends according to any one of claims 1 to 7.

Images