Methods for maintaining a plurality of similar components for aircraft as well as systems

Abstract

Method (1) for maintaining a plurality of identical components (2) for aircraft (3), wherein each of the identical components (2) is identifiable by a unique identification code (4), wherein the method (1) is characterized by the following process steps: a) Carrying out maintenance measures on at least a subset of the similar components (2), wherein carrying out a maintenance measure includes identifying possible fault conditions (7) and, if a fault condition (7) is present, assigning the identified fault condition (7) to a condition category (8) from a plurality of predefined condition categories (8); b) Storing the state categories (8) thus determined, each linked to the identification code (4) of the component (2) in which the fault state (7) was detected, and linked to a timestamp (9) indicating the time of detection of the fault state (7), as a fault data record (10); c) Creating a calculation model (11) that is set up to determine, on the basis of a current subset (12) of an operational data set (6), the probability (14) of the presence and / or future occurrence of at least one of the predefined state categories (8), individualized for individual components (2) of the similar components (2), wherein the operational data set (6) comprises measured values ​​(5) with which the operational state can be characterized individually for individual components (2) of the similar components (2), wherein within the operational data set (6) the measured values ​​(5) are each linked with the identification code (4) of the corresponding component (2) and with a timestamp (9) assigned to the respective measured value (5), wherein the creation of the calculation model (11) is based on a historical subset (13) of the operational data set (6) and on the basis of the fault data set (10),where, within the historical subset (13), the measured values ​​(5) are linked to timestamps (9) that precede the timestamps (9) to which the measured values ​​(5) of the current subset (12) are linked; d) Determining the probability (14) of the existence and / or future occurrence of at least one of the predefined state categories (8) individualized for individual components (2) of the similar components (2) on the basis of a current subset (12) of the operational data set (6) by processing the current subset (12) through the created calculation model (11); e) Generating an instruction (15) for carrying out a maintenance measure individualized for individual components (2) of the similar components (2) on the basis of the determined probability (14) of the presence and / or future occurrence of one of the predefined condition categories (8); and f) Performing a maintenance measure on one of the similar components (2) based on the generated instruction (15).

Description

[0001] The present invention relates to a method for maintaining a plurality of identical components for aircraft having the features of the preamble of claim 1 and to a system having the features of the preamble of claim 15.

[0002] Various approaches to the maintenance of aircraft components are known from the state of the art. A maintenance event can, for example, involve only the inspection of a component, the replacement of certain operating materials or subcomponents, the repair of the component, or the replacement of the entire component. A fundamental distinction is made between planned and unplanned maintenance. Planned maintenance events can be carried out at predetermined intervals, such as monthly or annually, or after reaching a predefined number of flight hours or flight cycles. The intervals for such planned maintenance events are usually chosen so that any technical malfunctions that occur do not disrupt operations. These are distinct from unplanned maintenance events, which must be performed due to an unexpected technical malfunction.Since unplanned maintenance events disrupt flight operations, airlines and maintenance companies aim to reduce the proportion of unplanned maintenance events.

[0003] For this purpose, state-of-the-art approaches exist for monitoring the condition of components. By evaluating specific measured values, attempts are made to determine the technical condition of the component in real time (diagnosis) or to predict the remaining operating time of a component (prognosis). These approaches allow suitable maintenance measures to be initiated even before a technical fault occurs, or the intervals between two maintenance measures to be optimally utilized. In this way, condition monitoring can make flight operations safer and more cost-efficient.

[0004] US Patent 10 228 687 B2 discloses a method for diagnosing a fault in an air-conditioning unit, referred to in English as an "air-conditioning pack". This involves comparing data from such an air-conditioning unit with threshold values.

[0005] US patent 2022 / 0026895A1 discloses a data processing system for generating predictive maintenance models.

[0006] US patent 2022 / 0281618A1 discloses a procedure for predictive maintenance of components by, among other things, creating a reliability curve for an aircraft based on historical modifications.

[0007] US Patent 2018 / 0349532A1 discloses a method for evaluating a part, whereby data are determined that represent a damage ranking model and a cumulative damage model for the part.

[0008] US patent 2017 / 0 193 372 A1 discloses a method, a device, and a system for evaluating the condition of a vehicle component. A computer system generates a digital forecast for the vehicle component; the digital forecast predicts whether maintenance should be performed on the component.

[0009] US Patent 2023 / 0061096A1 discloses a method for aircraft maintenance comprising loading a large number of unstructured aircraft component records into computer memory. The large number of unstructured aircraft component records are then transferred from computer memory to a natural language processing (NLP) model.

[0010] US patent 2003 / 0114965A1 discloses a method and system for an improved vehicle monitoring system. Machine learning and data mining technologies are applied to data collected from a large number of vehicles.

[0011] EP 3 379 359 B1 discloses a computer-implemented method for detecting a malfunction of a shut-off valve.

[0012] EP 3 608 744 A1 discloses a method for identifying linked events in an aircraft.

[0013] The purpose of this application is to provide an improved method for maintaining a plurality of similar components for aircraft, as well as a correspondingly improved system.

[0014] The problem is solved by the features of the independent claims. Further preferred embodiments of the invention can be found in the dependent claims, the figures, and the accompanying description.

[0015] According to a first aspect of this application, the problem is solved by a method for maintaining a plurality of identical components for aircraft, wherein each of the identical components is identifiable by a unique identification code, and wherein the method comprises the following procedural steps: Step a) Performing maintenance measures on at least a subset of the similar components, wherein performing a maintenance measure includes identifying possible fault conditions and, if a fault condition is present, assigning the identified fault condition to a condition category from a plurality of predefined condition categories. Step b) Storing the state categories thus determined, each linked to the identification code of the component in which the fault condition was detected, and linked to a timestamp that indicates the time of detection of the fault condition, as a fault data record. Step c) Creating a calculation model that is configured to determine, individually for individual components of the same type, the probability of the presence and / or future occurrence of at least one of the predefined state categories based on a current subset of an operational data set, wherein the operational data set comprises measured values ​​with which the operational state can be characterized individually for individual components of the same type, wherein within the operational data set the measured values ​​are each linked to the identification code of the corresponding component and to a timestamp assigned to the respective measured value, wherein the calculation model is created based on a historical subset of the operational data set and on the basis of the fault data set, wherein within the historical subset the measured values ​​are linked to timestamps.which precede the timestamps to which the measured values ​​of the current subset are linked. Preferably, the timestamps are assigned to the measured values ​​when the measured values ​​are first stored on a data storage device. Step d) Determining the probability of the existence and / or future occurrence of at least one of the predefined state categories, individualized for individual components of the similar components, based on a current subset of the operational data set, by processing the current subset through the created calculation model. Step e) Generating an instruction for carrying out a maintenance measure individualized for individual components of the similar components based on the determined probability of the presence and / or future occurrence of one of the predefined condition categories. Step f) Perform a maintenance measure on one of the similar components based on the generated instruction.

[0016] The invention recognizes that the information contained in the fault data set contributes significantly to the creation of a high-quality calculation model. By considering this information together with the information from the operational data set, the calculation model can calculate the probability of the presence and / or future occurrence of a specific condition category with a sufficiently high degree of accuracy. The calculation model can thus be used for diagnosing and / or predicting the presence or occurrence of condition categories, thereby making component maintenance more efficient. Based on this condition category, for example, it is possible to deduce the underlying fault condition or at least narrow down the possible fault conditions.

[0017] By using the historical subset of the operational data and the fault data for creating the calculation model, it becomes possible to establish relationships between the determined condition categories and the recorded measurements. The use of condition categories, which serve as a kind of "label" to categorize the various fault states in a practical way for maintenance operations, is particularly helpful in this regard. The use of predefined condition categories facilitates their machine processing, allowing them to be used efficiently for creating the calculation model.

[0018] By examining multiple components of the same type and documenting the observed fault conditions by assigning them to predefined condition categories, causal relationships between changes in one or more measured values ​​and the presence of a fault condition underlying a given condition category can be identified and considered when developing the action plan. Furthermore, the identical components are also operated in a large number of aircraft, preferably of the same type, thus providing a sufficiently large dataset for the development of the calculation model. Therefore, the proposed calculation model is capable of calculating the probability of the presence and / or future occurrence of a specific condition category with improved accuracy.By carrying out the process steps a) to f), the overall system reliability of the similar components is improved.

[0019] For the purposes of this application, a work instruction may encompass all activities typically involved in the maintenance of the relevant component; this may include, for example, replacing the entire component or subcomponents; it may also include, for example, repairing subcomponents and / or performing adjustment and / or setting operations. The work instruction may be issued, for example, in text and / or speech form. The work instruction may, for example, refer to measures described in a manual provided by the manufacturer of the aircraft or component. If the aircraft is an airplane, for example, reference may be made to an Aircraft Maintenance Manual (AMM) and / or a Component Maintenance Manual (CMM).

[0020] By issuing customized instructions for a specific component, time-consuming troubleshooting can be avoided. The instructions for a component with a specific identification code are preferably generated based on a current subset of the operational data set, specifically one generated by that component. It goes without saying that this current subset of the operational data set can be continuously updated and / or expanded during operation.

[0021] Preferably, the operational data set comprises a plurality of operational data records. Such an operational data record includes at least one measured value, an identification code, and a timestamp. Naturally, such an operational data record can also include further data fields, such as the registration number of the aircraft in which the component was operated when the measured values ​​were generated. Depending on the timestamp of the operational data records, they are assigned to the historical subset or the current subset of the operational data set. The operational data set can be continuously expanded with new operational data records, which are then provided to the calculation model as input data. Operational data records from the current subset of the operational data set can, over time, become part of the historical subset of the operational data set, thus enabling improvements to the calculation model.

[0022] Similarly, the error database preferably comprises multiple error records. Each such error record includes at least one status category, an identification code, and a timestamp. Naturally, such an error record can also include further data fields. The error database can, for example, be updated and / or expanded over time with additional error records.

[0023] The operational data and / or fault data records can be distributed across various physical storage devices. For example, the operational data records can be distributed across the storage devices of different aircraft and / or across the storage devices of stationary computers at one or more ground stations.

[0024] For the purposes of this application, "similar components" means components that are identical in construction or substantially identical in construction. Identical components are interchangeable for installation in aircraft. Essentially identical components include, in particular, components that are the same in terms of their technical function and performance data, but have different interfaces, such as different connections, for installation in the aircraft.

[0025] It goes without saying that even after the operational data set has been divided into a historical and a current subset, the link between measured values, timestamps and identification codes remains.

[0026] The characterization of the operating state within the meaning of this application can be carried out using a wide variety of operating parameters. An operating parameter can, for example, be a state (e.g., "on"-"off", "open"-"closed"), but also a physical quantity, such as pressure, temperature, or mass flow rate.

[0027] Preferably, the aircraft are of the same type, for example Airbus A320 aircraft, i.e., essentially identical aircraft.

[0028] Preferably, the generation of the operational data set comprises the following process steps: Step A) Operating a large number of aircraft, each containing at least one of the identical components. Step B) Generating the measured values ​​with which the operating state of individual components of the same type can be characterized when they are in operation in the respective aircraft. Step C) Saving the generated measured values ​​linked with the identification codes of the corresponding components and with the corresponding timestamps as an operational data set.

[0029] Steps A) to C) enable the operational data set to be generated in such a way that the data is suitable both for creating the calculation model and as input data for the calculation model.

[0030] It is further proposed that operational data generated by one of the aircraft during operation be transmitted to a ground station, with the computational model being executed on a computer at the ground station. The operational data is preferably transmitted during flight, for example, in the form of an Aircraft Condition Monitoring System (ACMS) report, and / or the transmission occurs after landing, for example, in the form of full flight data. Of course, alternative transmission methods are also possible, such as using an internet connection. The ground station can also comprise several spatially separated subunits.Preferably, the ground station comprises one or more data storage devices with a database on which the current subset of the operational data set, the historical subset of the operational data set and the error data set are stored.

[0031] It is further proposed that at least two identical components be installed in one of the aircraft, wherein a measured value characterizing the operating state of the first component in the aircraft is compared with a measured value characterizing the operating state of the second component within the same aircraft, wherein the comparison of these measured values ​​is an additional component of the historical subset of the operational data set and / or the current subset of the operational data set, and / or the comparison of these measured values ​​is performed by the computational model. The two components of an aircraft are preferably operated in parallel. Preferably, the comparison includes calculating the difference between the two measured values, which have the same or approximately the same timestamp, taking into account the sign of the determined difference value.The two measured values ​​being compared preferably originate from corresponding sensors on two identical components installed within the same aircraft. Comparing two measured values ​​from two identical components installed in the same aircraft improves the data basis for determining condition categories. For example, a simultaneous increase in a specific measured value on both components of an aircraft may indicate that this increase is not due to a fault condition, but rather to a change in environmental conditions or the flight condition. This, in turn, improves the overall quality of the results from the computational model.In order to assign the two components, which are intended for use in one and the same aircraft, to the corresponding aircraft, the operational data preferably includes another type of identification code with which the aircraft in which the component is installed during data generation can be identified. This additional type of identification code is then also linked to the associated measured values.

[0032] Preferably, an additional state category is provided that characterizes an intact system state of one of the similar components, wherein in process step a) the assignment to this state category takes place if a fault state is not present. This allows the absence of a fault state to also be used as information for creating the calculation model.

[0033] Preferably, step d) is repeated with newly generated data that is added to or replaces the current subset of the operational data, so that the probability of the presence and / or future occurrence of at least one of the predefined condition categories is updated and provided at predetermined intervals. This allows the probability of the presence and / or future occurrence of at least one of the predefined condition categories to be continuously determined. If necessary, a countermeasure in the form of an instruction for a maintenance measure is created.By continuously determining these probabilities, the maintenance measure can be carried out in a particularly advantageous way before the occurrence of the fault condition underlying the condition category; the maintenance measure can thus be carried out as a preventive maintenance measure.

[0034] It is further proposed that step a) comprises the following sub-steps: performing maintenance on at least a subset of the similar components in an installed state, wherein the respective component is installed in one of the aircraft; and performing maintenance on at least a subset of the similar components in a removed state, wherein the respective component is removed from one of the aircraft. It has been shown that significantly better results can be achieved by considering all available maintenance data concerning the similar components for the creation of the calculation model. Maintenance of components in the installed state on the aircraft is generally limited.Maintenance measures requiring a higher level of detail and / or disassembly than when the component is installed can only be performed in a suitably equipped workshop. Different fault conditions are frequently diagnosed when the component is installed compared to when it is removed. Therefore, the database can be improved by using data generated during maintenance measures in both the installed and removed states. For the purposes of this application, a removed state is defined as a state in which the component is neither structurally nor functionally connected to an aircraft. In other words, the component can be transported independently of the aircraft when removed.

[0035] It is further proposed that each of the identical components includes at least one sensor that generates measured values ​​during operation to characterize the operating state of that component; and / or that each of the aircraft includes at least one sensor that generates measured values ​​during operation to characterize the operating state of one of the identical components installed in the aircraft. This approach allows for particularly efficient generation of the measured values ​​required for the computational model. It is also not essential that the sensors be part of the component itself; for example, sensors intended for monitoring adjacent components in the aircraft are also suitable. Various sensor types are possible, such as sensors for determining temperature, the position of an actuator, atmospheric pressure, and / or flow rate.

[0036] Preferably, the calculation model is configured to determine a general system state individually for each component of the same type. For example, an instruction for action can only be issued when a component reaches a critical general system state. The general system state can be determined, for instance, based on the calculated probability of the presence and / or future occurrence of at least one of the predefined state categories.

[0037] Preferably, the general system state is represented by a numerical value, wherein the computational model is configured to compare the numerical value representing the general system state with a first and a second threshold, wherein a signal is output if the numerical value is between the first and second thresholds, and wherein the generation of the action instruction according to step e) is triggered if the numerical value exceeds the second threshold. The numerical value for representing the general system state offers the advantage of being easy and efficient to process. The thresholds are, for example, determined empirically.

[0038] Preferably, in step c), the computational model is at least partially created and / or adapted using machine learning, with the historical subset of the operational data and the fault data being used as training data for the machine learning. It has been shown that machine learning is particularly efficient at determining causal relationships between the fault data and the historical subset of the operational data. This improves the quality of the computational model.

[0039] It has proven advantageous to implement machine learning through the use of so-called boosting algorithms. Integrating boosting algorithms can improve the performance of diagnosing and / or predicting the presence or occurrence of state categories. These algorithms have been shown to improve prediction accuracy and overall performance in various machine learning tasks. Their adaptability and versatility, in particular, make boosting algorithms suitable for the application described here. Of particular note in the context of this application is their ability to skillfully manage complex data relationships and mitigate noise.

[0040] It is further proposed that the identical components each comprise an air conditioning unit, including a cooling turbine, at least one, preferably four, heat exchangers, and a flow control valve. An air conditioning unit, also known as an Air Conditioning Pack (ACP), can, in the event of a malfunction, lead to operational limitations, uncomfortably high cabin temperatures, or even smoke development. For this reason, the proposed method has particularly advantageous effects when applied to such an air conditioning unit. The component referred to here as a cooling turbine is technically called an Air Cycle Machine (ACM); in addition to a turbine, it also includes a compressor.

[0041] Preferably, the majority of predefined condition categories includes one or more of the following: defective cooling turbine, defective heat exchanger, and / or defective flow control valve. Considering at least one of these condition categories can already achieve a significant improvement in the system reliability of air conditioning systems.

[0042] According to a second aspect of this application, the aforementioned problem is solved by a system with a fleet comprising a plurality of aircraft and a plurality of identical components, in particular a plurality of identical air conditioning units comprising a cooling turbine, at least one, preferably four, heat exchangers and at least one flow control valve, wherein at least one of the identical components is installed in each of the aircraft, wherein the maintenance of at least a subset of the identical components is carried out using the method according to the first aspect of this application, optionally taking into account the developments described above, wherein the system comprises a computer program product configured to carry out the method steps c) to e) according to the first aspect of this application.Regarding the technical effects and advantages associated with the proposed system according to the second aspect of this application, reference is made to the preceding explanations in connection with the method according to the first aspect of this application. Preferably, the computer program product is loaded into an internal memory of a computer, which is, for example, part of the ground station, and is executed by this computer.

[0043] The invention is explained below with reference to preferred embodiments and the accompanying figures. These figures show: Fig. 1 a schematic representation of a component; Fig. 2 a first schematic representation of a procedure, Fig. 3 a second schematic representation of a procedure; Fig. 4 a schematic representation of a comparison of measured values; Fig. 5. a temporal progression of a general health status of a component; Fig. 6. Time courses of the probability of the existence of different state categories; and Fig. 7 a system.

[0044] Fig. Figure 1 shows a component 2 for an aircraft 3 (see Figure 7), namely an air conditioning unit 23 for an aircraft, also known as an Air Conditioning Pack (ACP).

[0045] The air conditioning unit 23 comprises a flow control valve 26, four heat exchangers 25a to 25d, and a cooling turbine 24, also known as an air cycle machine (ACM). The cooling turbine 24 in turn comprises a compressor 32 and a turbine 33.

[0046] The air conditioning unit 23 cools bleed air 30 from an engine (not shown) so that it can be pre-cooled before entering the cabin of an aircraft 3. The flow control valve 26 regulates the amount of bleed air 30 supplied to the air conditioning unit 23 from the respective engine. The air conditioning unit 23 also includes a flow channel 31 through which cold ram air 34 flows.

[0047] The bleed air 30 is drawn from a hot, high-pressure area of ​​the engine and flows through the flow control valve 26 and then over a first heat exchanger 25a, which is cooled by the relatively cold ram air 34 from the flow channel 31. From the first heat exchanger 25a, a portion of the cooled bleed air 30 flows into the compressor 32 of the cooling turbine 24, so that the temperature is increased again by compression in the compressor 32. The temperature difference between the bleed air 30 and the ram air 34 at the heat exchanger 25b is thus increased in order to transfer more heat energy to the ram air 34 flowing past it. From the heat exchanger 25b, the bleed air 30 then flows through the two further heat exchangers 25c and 25d, transferring heat energy, into the turbine 33 of the cooling turbine 24; the heat exchanger 25c is passed through twice on the way to the turbine 33.Downstream of the heat exchanger 25d, a water separator 42 is provided for separating condensed water. The expansion in the turbine 33 causes further significant cooling of the bleed air 30. After exiting the turbine 33, the bleed air 30 flows through the heat exchanger 25d once more to absorb thermal energy. The bleed air 30 then flows at approximately room temperature through an interface 36 into a flow line 35 of the aircraft 3. A bypass valve 37 is also shown, which allows adjustment of the proportion of the bleed air 30 that bypasses the compressor 32; this proportion then flows via a bypass 38 directly to the outlet side of the turbine 32. The amount of ram air 34 flowing into the flow channel 31 is defined by the position of a flap 39, the position of which can be adjusted via an actuator 40. Furthermore, it shows . Fig. Figure 1 on the right side schematically shows another component 2 in the form of an air conditioning unit 23, which is connected to another airflow line 35 of the same aircraft 3. The design and function of the second air conditioning unit 23, shown only schematically on the right side, correspond to the design and function of the air conditioning unit 23 shown in detail on the left side.

[0048] Furthermore, the Fig. 1 various sensors 18a-18h, with which measured values ​​5 are obtained during operation (see for example) Fig. 2 and Fig. 4) to characterize the operating state of the air conditioning unit 23. The valve position of the flow control valve 26 is determined as measured value 5 using a sensor 18a. The differential pressure of the bleed air 30 in a flow channel upstream of the flow control valve 26 relative to the environment 41 is determined as measured value 5 using another sensor 18b. The pressure within the flow line upstream of the flow control valve 26 is determined as measured value 5 using a sensor 18c, i.e., the pressure applied to the inlet side of the air conditioning unit 23. Furthermore, the valve position of the bypass valve 37 is determined as measured value 5 using a sensor 18d. Finally, the outlet temperature of the bleed air 30 from the compressor 32 is determined as measured value 5 using sensor 18e.A further sensor 18f is provided, which determines the position of the flap 39 for controlling the amount of incoming stagnant air 34 as measured value 5. The temperature downstream of the water separator 42 is determined as measured value 5 using sensor 18g. Finally, the outlet temperature of the air conditioning unit 23 is determined as measured value 5 using sensor 18h; that is, the temperature at which the cooled bleed air 30 is discharged into the flow line 35.

[0049] Air conditioning systems 23 as described above are already known from the prior art.

[0050] The measured values ​​5 obtained by means of sensors 18a to 18h are used as input data for the procedure 1 described below (see Fig. 2, Fig. 3 and Fig. 7) used for the maintenance of a majority of the air conditioning units 23. The following description of this method 1 using the air conditioning unit 23 of an aircraft as an example is only to be understood as such; it can be applied equally to other similar components 2 of an aircraft.

[0051] Fig. Figure 2 schematically shows a method 1 for the maintenance of a plurality of air conditioning units 23, whose structure and function are already known from the Fig. 1 is known. Measured values ​​5a and 5b are shown as examples, which characterize the operating state individually for one of the air conditioning units 23. Measured values ​​5a and 5b are, for example, measured values ​​5, which are obtained from two of the sensors 8a up to 18h. Fig. 1 were determined. For clarity, only two measured values, 5a and 5b, are shown graphically here. The air conditioning unit 23 under consideration is assigned an identification code 4a. It is indicated graphically that such measured values ​​5a and 5b are also recorded for further air conditioning units 23 with the identification codes 4b and 4c. Naturally, in practice, the measured values ​​5a and 5b will be recorded for more than just three identical components 2 in the form of air conditioning units 23. The measured values ​​5a and 5b of the air conditioning units 23, together with their assigned identification codes 4a, 4b, and 4c, and with corresponding timestamps 9, which are assigned when the measured values ​​5a and 5b are saved, form an operational data set 6, which is divided into a historical subset 13 and a current subset 12.The current subset 12 of the operational data set 6 has more recent timestamps 9 than the historical subset 13 of the operational data set 6. In other words, the measured values ​​5a, 5b of the historical subset 13 of the operational data set 6 are older than the current subset 12 of the operational data set 6.

[0052] Furthermore, in Fig. Figure 2 schematically illustrates the creation of an error database 10. For this purpose, a process step a) is first carried out, the integration of which into process steps A)-C) and b)-f) is subsequently explained using the following: Fig. 3 will be explained in detail.

[0053] Procedure step a) comprises performing maintenance measures on several air conditioning units 23, wherein performing these maintenance measures includes a') determining possible fault conditions 7 and, if a fault condition 7 is present, a'') assigning the determined fault condition 7 to a condition category 8 from a plurality of predefined condition categories 8. In this embodiment, a catalog comprising a large number of fault conditions 8 for various fault conditions 7 is provided. For example, it may be determined during the performance of the maintenance measure that no fault is present; the absence of a fault is then also assigned to a specially provided condition category 8b, which characterizes an intact system state 46.The condition categories 8a and 8b are linked with corresponding timestamps 9 and identification codes 4a to 4c; they thus form the error data set 10.

[0054] Furthermore, it shows Fig. 2, that the historical subset 13 of the operational data set 6, together with the fault data set 10, constitute training data 43, which are used to create—and in this embodiment also to optimize—the calculation model 11. The training data 43 thus generated make it possible to determine relationships between changes in the measured values ​​5a, 5b and specific operating states. The arrow 47 indicates the time of detection of a specific fault state 7a during a maintenance measure in step a), which is assigned to state category 8a. The measured values ​​5a and 5b are plotted here as a function of time. It is evident that the function values ​​exhibit significant fluctuations before the detection of the fault state 7a.By combining the corresponding measured values ​​5a and 5b with the determined state category 8a, the calculation model 11 can be created in such a way that it can assign certain characteristics of the temporal progression of the measured values ​​5a and 5b to a specific state category 8. This applies in particular the larger the number of similar components 2 whose data are available for creating the operational data set 6 and the fault data set 10. Furthermore, an arrow 48 indicates the time of detection of an intact system state 46. It is evident that the measured values ​​5a and 5b do not exhibit any anomalies before the time indicated by the arrow 48. This information is used in the creation of the calculation model 11 to define the actual presence of state category 8b, intact system state 46.

[0055] The creation of the computational model 11 is carried out here using machine learning (ML) methods with the training data 43. In this embodiment, the machine learning methods also include the use of so-called boosting algorithms. In principle, however, alternative algorithms can also be used.

[0056] The calculation model 11, created on the basis of the historical subset 13 of the operational data set 6 and on the basis of the error data set 10, is designed to continuously process a current subset 13 of the operational data set 6. For example, the data of the current subset 13 can be provided to the calculation model 11 as input data. The calculation model 11 processes this input data in such a way that the probability 14 (see Fig. 6) for the presence - and in this embodiment also for the future occurrence - of at least one of the predefined state categories 8a, 8b can be determined individually for individual components 2 of the similar components 2.

[0057] Based on the determined probability 14, an action instruction 15 for carrying out a maintenance measure is generated for the affected air conditioning unit 23 using the calculation model 11. Based on this action instruction 15, the action instruction 15 proposed by the calculation model 11 is then carried out on the affected air conditioning unit 23 in a process step f). Furthermore, the calculation model 11 can also be configured to output the diagnosed or predicted condition category 8 on which the action instruction 15 is based. The maintenance personnel are thus aware of the category of the fault that is to be rectified by the action instruction 15. The calculation model 11 can be subjected to a validation 44.

[0058] Fig. Figure 3 schematically shows the process of the proposed procedure 1: The operational data set 6 is generated using the procedure steps A) to C).

[0059] In step A) a large number of aircraft 3 (see Fig. 7), in which two of the air conditioning units 23 are installed. In step B), measured values ​​5 are generated with which the operating state of individual air conditioning units 23 can be characterized when they are in operation in the respective aircraft 3. The generated measured values ​​5 are then linked in step C) with the identification codes 4 of the corresponding air conditioning units 23 and with the timestamps 9 and stored as an operational data set 6. Of course, the operational data set 6 can be continuously supplemented.

[0060] In step a), a maintenance measure is carried out on at least a subset of the air conditioning units 23, wherein the execution of a maintenance measure includes the identification of possible fault states 7. If a fault state 7 is identified, it is assigned to a corresponding state category 8 from a plurality of predefined state categories 8. Provided that the air conditioning unit 23 has an intact system state 46 (see Fig. 2) exhibits the corresponding condition category 8b (see Fig. 2) awarded.

[0061] In step b), the condition categories 8 determined over a longer period of time for a large number of air conditioning units 23 are linked to the identification code 4 of the air conditioning unit 23 in which the corresponding fault condition 7 was detected, and linked to a timestamp 9 that indicates the time of detection of the fault condition 7, and stored as fault data 10.

[0062] In step c), the calculation model 11 is created. It is set up to be individualized for individual air conditioning units 23 of the similar air conditioning units 23 based on the current subset 12 (see Fig. 2) of the operational data set 6 the probability 14 (see Fig. 6) to determine and output the presence and / or future occurrence of at least one of the predefined state categories 8. As explained above, the operational data set 6 comprises the measured values ​​5, which characterize the operating state individually for each air conditioning unit 23 of the similar air conditioning units 23. In the operational data set 6, the measured values ​​5 are each linked to the identification code 4 of the corresponding air conditioning unit 23 and to a timestamp 9 assigned to the measured values ​​5 when they are saved. The calculation model 11 is then created based on the historical subset 13 of the operational data set 12 and on the fault data set 10, i.e., based on the training data 43.

[0063] In step d), the probability 14 for the existence and / or future occurrence of at least one of the predefined state categories 8 is determined individually for individual air conditioning units 23 of the similar air conditioning units 23 on the basis of a current subset 12 of the operational data set 6 by processing the current subset 12 through the created calculation model 11.

[0064] In step e), the instruction 15 for carrying out a corresponding maintenance measure is individualized for individual air conditioning units 23 of the similar air conditioning units 23 on the basis of the determined probability 14 for the presence and / or future occurrence of one of the predefined condition categories 8.

[0065] Finally, in step f), a maintenance measure is carried out on one of the similar air conditioning units 23 based on the generated instruction 15.

[0066] Fig. Figure 4 schematically shows the comparison of corresponding measured values ​​5a to 5e from two different air conditioning units 23, which were installed simultaneously in the same aircraft 23. The measured values ​​5a to 5e were recorded for the first and second air conditioning units 23 respectively; since they are provided with timestamps 9, the measured values ​​5a to 5e can be plotted as a function of time. In the Fig. 4. The measured values ​​5a to 5e for the first and second air conditioning units 23 are displayed as separate trends in the corresponding diagram. By calculating the difference between the identical measured values ​​5 of the two air conditioning units 23 at defined times, a comparison of the measured values ​​5a to 5e of the two air conditioning units 23 can then be made. The signs are taken into account when calculating the difference.

[0067] Fig. Figure 5 shows a time course of a general system state 20 of an air conditioning system 23. The general system state 20 is a numerical value determined from the probabilities 14 for the presence of certain state categories 8, calculated by the computational model 11. The determination of the general system state 20 is also carried out using the computational model 11. A value of 0 indicates a very good general system state 20, while a value of 1.0 indicates a very poor general system state 20. Furthermore, the diagram shows two threshold values ​​21 and 22, which were determined empirically. If the general system state 20 exceeds the first threshold value 21 but remains below the second threshold value 22, the computational model 11 issues a signal in the form of a warning. This informs the maintenance personnel about a possible future fault condition 7.If the general system state 20 exceeds the second threshold 22, then the calculation model 11 triggers the generation of an instruction 15.

[0068] Fig. Figure 6 shows, as an example, the time course of the probabilities 14 for the occurrence of the following condition categories 8: defective cooling turbine 24, defective flow control valve 26 and defective heat exchanger 25. These failure probabilities are determined by the calculation model 11.

[0069] Finally, the Fig. 7 a system 100 with which the previously described method 1 can be applied. It comprises a fleet 53 with a plurality of aircraft 3 in the form of airplanes, each comprising two of the identical components 2 in the form of air conditioning units 23.

[0070] It is schematically depicted that the aircraft 3 generate data during operation, which is part of the operational data set 6. Within an operational data set, one or more measured values ​​5 are linked with a corresponding identification code 4 and a corresponding timestamp 9. A plurality of such operational data sets constitutes the operational data set 6. It goes without saying that the operational data set 6 does not have to be stored in a single physical location.

[0071] Furthermore, maintenance 51 is shown schematically, which is subdivided into aircraft maintenance 52 and component maintenance 49. In aircraft maintenance 52, maintenance measures on the air conditioning unit 23 can be carried out in an installed state 16; a condition category 8 determined in this way, together with the identification code 4 and a timestamp 9, then becomes part of the fault data set 10 as a fault record. In component maintenance 49, the maintenance measure on the air conditioning unit 23 is carried out in a removed state 17; the spatial separation of the air conditioning unit 23 from the aircraft 3 is also shown graphically. The condition category 8 determined in the removed state 17, together with the identification code 4 and a timestamp 9, also then becomes part of the fault data set 10 as a fault record.

[0072] The historical subset 13 of the operational data set 6 and the error data set 10 are used – as described above – to create and optimize the calculation model 11. The calculation model 11 is loaded as a computer program product 27 into the internal memory of a computer 29. The computer 29 is part of a ground station 28.

[0073] Furthermore, the Fig. 7 For example, consider an aircraft 3 of fleet 53, which is in flight operation. The aircraft generates measured values ​​5 with which the operating status can be individually characterized for both installed air conditioning units 23. This is done using the sensors 18 of the air conditioning units 23 themselves, but also, for example, using sensors 19 that are assigned to the aircraft 3 but are not part of the air conditioning units 23. The corresponding measured values ​​5 are each linked with an identification code 4 and a timestamp 9 as an operational data record and transmitted wirelessly to a ground station 28 during flight. Of course, transmission of the operational data records on the ground is also possible. The operational data records generated in this way are initially part of the current subset 12 of the operational data set 6. They are provided to the calculation model 11 as input data.Any action instruction derived therefrom 15 and / or the general system state (see . Fig. 5) and / or the probabilities 14 (see Fig. 6) are displayed to the maintenance personnel on a display device 50, for example a screen. When an instruction 15 is displayed, it is carried out by the maintenance personnel on the corresponding air conditioning unit 23.

Claims

[1] Method (1) for maintaining a plurality of identical components (2) for aircraft (3), wherein each of the identical components (2) is identifiable by a unique identification code (4), wherein the method (1) is characterized by the following process steps: a) Carrying out maintenance measures on at least a subset of the similar components (2), wherein carrying out a maintenance measure includes identifying possible fault conditions (7) and, if a fault condition (7) is present, assigning the identified fault condition (7) to a condition category (8) from a plurality of predefined condition categories (8); b) Storing the state categories (8) thus determined, each linked to the identification code (4) of the component (2) in which the fault state (7) was detected, and linked to a timestamp (9) indicating the time of detection of the fault state (7), as a fault data record (10); c) Creating a calculation model (11) that is set up to determine, on the basis of a current subset (12) of an operational data set (6), the probability (14) of the presence and / or future occurrence of at least one of the predefined state categories (8), individualized for individual components (2) of the similar components (2), wherein the operational data set (6) comprises measured values ​​(5) with which the operational state can be characterized individually for individual components (2) of the similar components (2), wherein within the operational data set (6) the measured values ​​(5) are each linked with the identification code (4) of the corresponding component (2) and with a timestamp (9) assigned to the respective measured value (5), wherein the creation of the calculation model (11) is based on a historical subset (13) of the operational data set (6) and on the basis of the fault data set (10),where, within the historical subset (13), the measured values ​​(5) are linked to timestamps (9) that precede the timestamps (9) to which the measured values ​​(5) of the current subset (12) are linked; d) Determining the probability (14) of the existence and / or future occurrence of at least one of the predefined state categories (8) individualized for individual components (2) of the similar components (2) on the basis of a current subset (12) of the operational data set (6) by processing the current subset (12) through the created calculation model (11); e) Generating an instruction (15) for carrying out a maintenance measure individualized for individual components (2) of the similar components (2) on the basis of the determined probability (14) of the presence and / or future occurrence of one of the predefined condition categories (8); and f) Performing a maintenance measure on one of the similar components (2) based on the generated instruction (15). [2] Method according to claim 1, characterized by , that the creation of the operational data set (6) comprises the following procedural steps: A) Operating a plurality of aircraft (3), each of which incorporates at least one of the identical components (2); and B) Generating the measured values ​​(5) with which the operating state of individual components (2) of the same type can be characterized when they are in operation in the respective aircraft (3); C) Storing the generated measured values ​​(5) linked with the identification codes (4) of the corresponding components (2) and with the corresponding timestamps (9) as an operational data set (6). [3] Method according to claim 2, characterized by , that - data generated by one of the aircraft (3) during operation from the operational data set (6) are transmitted to a ground station (28), and wherein - the calculation model (11) is executed on a computer (29) at the ground station (28). [4] Method (1) according to one of claims 2 or 3, characterized by , that - at least two of the identical components (2) are installed in one of the aircraft (3), wherein - a measured value (5) that characterizes the operating state of the first component in the aircraft (3) is compared with a measured value (5) that characterizes the operating state of the second component (2) within the same aircraft (3), - wherein the comparison of these measured values ​​(5) is an additional component of the historical subset (13) of the operational data set (6) and / or the current subset (12) of the operational data set (6), and / or the comparison of these measured values ​​(5) is carried out by the calculation model (11). [5] Method (1) according to any of the preceding claims, characterized by , that - an additional state category (8) is provided which indicates an intact system state (46) of one of the similar components (2), wherein - in process step a) if there is no fault condition (7) the assignment to this state category (8) takes place. [6] Method (1) according to any of the preceding claims, characterized by , that - step d) is repeated with newly generated data that is added to or replaces the current subset (12) of the operational data set (6), so that the probability (14) of the presence and / or future occurrence of at least one of the predefined state categories (8) is updated and provided at specified time intervals. [7] Method (1) according to any of the preceding claims, characterized by , that step a) comprises the following sub-steps: - Performing maintenance measures on at least a subset of the identical components (2) in an installation state (16), wherein in the installation state (16) the respective component (2) is installed in one of the aircraft (3); and - Performing maintenance measures on at least a subset of the similar components (2) in a disassembled state (17), wherein in the disassembled state (17) the respective component (2) is removed from one of the aircraft (3). [8] Method (1) according to any of the preceding claims, characterized by , that - each of the identical components (2) comprises at least one sensor (18) with which measured values ​​(5) are generated during operation to characterize the operating state of that component (2); and / or - each of the aircraft (3) includes at least one sensor (19) which, during operation, generates measured values ​​(5) to characterize the operating state of one of the components (2) of the same type installed in the aircraft (3). [9] Method (1) according to any of the preceding claims, characterized by , that - the calculation model (11) is set up to determine a general system state (20) individualized for individual components (2) of the similar components (2). [10] Method (1) according to claim 9, characterized by , that - the general system state (20) is determined on the basis of the calculated probability (14) for the presence and / or future occurrence of at least one of the predefined state categories (8). [11] Method (1) according to one of claims 9 or 10, characterized by , that - the general system state (20) is represented by a numerical value, where - the calculation model (11) is set up to compare the numerical value representing the general system state (20) with a first and a second threshold (21, 22), wherein - a signal is output when a numerical value is between the first threshold (21) and the second threshold (22), and where - if the numerical value exceeds the second threshold (22), the generation of the instruction (15) according to step e) is triggered. [12] Method (1) according to any of the preceding claims, characterized by , that - the computational model (11) in step c) is at least partially created and / or adapted by machine learning (ML), wherein - the historical subset (13) of the operational data set (6) and the error data set (10) are used as training data (43) for machine learning (ML). [13] Method (1) according to any of the preceding claims, characterized by , that - the identical components (2) each comprise an air conditioning unit (23) comprising a cooling turbine (24), at least one, preferably four, heat exchangers (25) and a flow control valve (26). [14] Method (1) according to claim 13, characterized by , that - the majority of predefined condition categories (8) includes one or more of the following condition categories (8): faulty cooling turbine (24), faulty heat exchanger (25) and / or faulty flow control valve (26). [15] System (100) comprising a fleet of aircraft (3) and a plurality of identical components (2), in particular a plurality of identical air conditioning units (23) comprising a cooling turbine (24), at least one, preferably four, heat exchangers (25) and at least one flow control valve (26), wherein at least one of the identical components (2) is installed in each of the aircraft (3), characterized by , that - the maintenance of at least a subset of the identical components (2) is carried out using the method (1) according to one of the preceding claims, wherein - the system (100) comprises a computer program product (27) configured to perform the process steps c) to e) according to claim 1.

Images