Helicopter abnormal dynamic monitoring method based on working condition

Abstract

This invention discloses a helicopter anomaly dynamic monitoring method based on operating conditions, comprising the following steps: selecting a sample set of flight parameters; training an operating condition clustering model based on the sample set; training a monitoring feature mapping network for a single operating condition sample set until the difference between the generated monitoring feature value and the true value is minimized; calculating the residual X of healthy samples. e0 and X e0 mean μ e and variance σ e ; Perform time segmentation of the monitoring features for abnormal samples X to obtain the segmented monitoring features X under each time window. r Input the operating parameters of the abnormal sample X at the corresponding time into the operating condition clustering model to obtain the monitoring characteristics X of the abnormal sample. g Corresponding residual X e and the monitoring characteristics X of abnormal samples g mean μ g and variance σ g Based on segmentation monitoring feature X r Generate monitoring feature X g Residual X e and healthy sample residuals X e0 The relationship between the two, and the calculation of segmentation monitoring features X r The upper control limit, center line, and lower control limit.

Description

Technical Field

[0001] This invention belongs to the field of fault diagnosis and health management, and more specifically, relates to a method for abnormal dynamic monitoring of helicopters. Background Technology

[0002] Helicopters possess numerous advantages, including vertical takeoff and landing, no runway required, and the ability to hover in the air, making them play a vital role in both military and civilian applications. Reliability, maintainability, safety, testability, and supportability throughout a helicopter's entire lifespan are prerequisites for its continued airworthiness; therefore, the widespread adoption of helicopter HUMS (Helicopter Utility Management System) is an inevitable trend.

[0003] To assess the performance of the helicopter and its components during use and ensure safe operation, the HUMS system manufacturer sets status indication thresholds for each extracted characteristic parameter. These thresholds reflect the ideal operating values ​​of the target characteristic parameters under specified conditions when performing their intended functions. Exceeding the threshold triggers an alarm, requiring maintenance of the corresponding component; conversely, values ​​below the threshold indicate that the monitored component remains in good condition. The rationality and reliability of the monitoring benchmark thresholds are crucial for the accuracy of helicopter condition assessment.

[0004] On the one hand, inappropriate threshold settings can lead to false alarms, which can be mainly divided into missed detections and false positives. During helicopter operation, the HUMS system is more likely to issue false positive alarms than missed detection alarms. This is because when designing the HUMS, manufacturers try to avoid missed detections and pursue a high detection rate, hoping to reduce the probability of monitored component failures by performing more maintenance and inspection work. However, too many false positive alarms will place a huge burden on maintenance and inspection work.

[0005] On the other hand, helicopters require multiple takeoffs, landings, and hoverings during flight missions. They encounter various harsh environments during normal flight, necessitating constant adjustments to their flight status to adapt. This leads to highly variable operating conditions in the helicopter's transmission system, potentially causing trend changes in its characteristics and even triggering alarms. However, from a usage and maintenance perspective, these so-called alarm points are unreasonable. Because once the helicopter leaves a specific operating condition, the fault alarm automatically clears, and the alarm does not affect the helicopter's continued airworthiness. Therefore, using static state indication thresholds as the monitoring benchmark for changes in the target equipment's operating status will result in unreasonable and inaccurate monitoring and evaluation results.

[0006] Therefore, given the various flight conditions affecting helicopters, fixed monitoring thresholds do not consider the impact of these conditions on monitoring characteristics, leading to unsatisfactory monitoring results. Research is needed on a condition-based dynamic monitoring method for helicopter anomalies. Summary of the Invention

[0007] The purpose of this invention is to provide a helicopter anomaly dynamic monitoring method based on operating conditions. Under the real-time changing operating conditions, this method enables dynamic anomaly monitoring by providing monitoring feature control limits that change synchronously with flight parameters and flight conditions. This effectively improves the false alarm and missed detection problems caused by fixed thresholds. Furthermore, this method can be applied to real-time helicopter status monitoring, providing a basis for helicopter maintenance management decisions.

[0008] To achieve the above objectives, the specific implementation steps of the helicopter anomaly dynamic monitoring method based on operating conditions of the present invention are as follows:

[0009] Select a sample set of flight parameters;

[0010] Train the working condition clustering model based on the sample set;

[0011] Training of the monitoring feature mapping network for single-condition sample sets is performed until the difference between the generated monitoring feature value and the true value is minimized.

[0012] For the trained single-condition sample set monitoring feature mapping network, input healthy samples under the corresponding conditions, and calculate the healthy sample residual X. e0 and X e0 mean μ e and variance σ e ;

[0013] Perform time segmentation of the monitoring features for the abnormal sample X to obtain the segmented monitoring features X under each time window. r ;

[0014] Input the operating condition parameters of the abnormal sample X at the corresponding time into the operating condition clustering model to obtain the monitoring characteristics X of the abnormal sample. g Corresponding residual X e and the monitoring characteristics X of abnormal samples g mean μ g and variance σ g

[0015] Based on segmentation monitoring feature X r Generate monitoring feature X g Residual X e and healthy sample residuals X e0 The relationship between the two, and the calculation of segmentation monitoring features X r The upper control limit, center line, and lower control limit.

[0016] In a further embodiment of the present invention, the sample set of flight parameters includes at least:

[0017] The training includes maintenance manuals, system structure and principles, expert experience, a flight parameter sample set with selected key operating condition flight parameters as input and corresponding operating condition labels as output, and a flight operating condition clustering model trained based on a Gaussian mixture model.

[0018] In a further embodiment of the present invention, the training of the single-condition sample set monitoring feature mapping network takes the condition flight parameters as input and the corresponding health monitoring features as output.

[0019] The health monitoring features are those of the directly monitored object or features of the monitored object after feature extraction.

[0020] The helicopter's flight parameters and health monitoring characteristics are consistent over time.

[0021] In a further embodiment of the present invention, the time segmentation of monitoring features for abnormal sample X specifically involves selecting a certain window length w. L The monitoring features are segmented using a sliding time window with a step size Δw, so that each time window has segmented monitoring features X. r .

[0022] In a further embodiment of the invention, the window length w of the sliding time window L The step size Δw is inversely proportional to the timeliness requirement of anomaly monitoring, and the window length w of the sliding time window... L The step size Δw is inversely proportional to the fluctuation of the monitoring characteristics.

[0023] In a further embodiment of the present invention, the upper control limit, the center line, and the lower control limit are calculated according to the following formula:

[0024] X = X + X

[0025] X r =X g +X e

[0026] μ r0 =E(X) g +X e )=f(μ g ,μ e )

[0027]

[0028] UCL = μ r0 +3σ r

[0029] CL = μ r0

[0030] LCL = μ r0 -3σ r0.

[0031] Compared with the prior art, the advantages of this invention are that the dynamic control limit takes into account the influence of flight parameters and flight conditions on the monitoring characteristics, effectively improving the false alarm and missed detection problems caused by fixed thresholds. It can be used not only for the evaluation of flight sorties on the ground, but also for the real-time status monitoring of helicopters, providing a basis for decision-making for helicopter maintenance and management.

[0032] To address the issue that helicopters experience multiple flight conditions during a single flight, fixed monitoring thresholds fail to account for the impact of these conditions on monitoring characteristics, leading to unsatisfactory monitoring results, including numerous false alarms and missed detections. This invention, based on flight condition identification training and single-condition health monitoring feature mapping network training, utilizes a sliding window approach. Within each window, flight status is identified and matched to a corresponding monitoring feature mapping network, resulting in generated monitoring features. These features, combined with their statistical parameters and residual statistical parameters, are used to calculate segmented monitoring feature control limits within the window. As the window slides, the control limits change synchronously with flight parameters and flight conditions, enabling dynamic anomaly monitoring. The method provided by this invention generates dynamic monitoring feature control limits based on flight conditions, effectively improving the false alarm and missed detection problems caused by fixed thresholds. Furthermore, this method can be applied to real-time helicopter status monitoring, providing a basis for helicopter maintenance and management decisions. Attached Figure Description

[0033] Figure 1 This is a flowchart of a helicopter anomaly dynamic monitoring method based on operating conditions provided in one embodiment of the present invention.

[0034] Figure 2 The window length w is provided in one embodiment of the present invention. L A schematic diagram of dynamic monitoring of fault samples with a step size of 10 and a step size of Δw = 0. Detailed Implementation

[0035] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to the accompanying drawings, but this should not be construed as limiting the scope of protection of the present invention. It should be noted that, unless otherwise specified, the embodiments and various methods described in this application can be combined with each other. The accompanying drawings described below are merely some embodiments of the present invention; those skilled in the art can obtain other drawings based on these drawings without any creative effort.

[0036] See Figure 1 This study proposes a condition-based dynamic monitoring method for helicopter anomalies, which includes:

[0037] S100: Based on the helicopter's historical flight parameter set, select key flight parameters that reflect the flight conditions according to certain criteria. These criteria include, but are not limited to, maintenance manuals, system composition and principles, and expert experience. Obtain the condition labels based on the flight mission records, that is, generate a flight parameter sample set for each condition. The selected key flight parameters are the input, and the corresponding condition labels are the output. Based on the flight parameter sample set for each condition, train the flight condition clustering model using a Gaussian mixture model (GMM) to enable the flight condition identification model to correspond to the actual conditions based on the condition labels obtained from clustering.

[0038] S200: Under each single operating condition, the helicopter operating condition flight parameters are used as input and the corresponding helicopter health monitoring features are used as output to train the single operating condition monitoring feature mapping network. The helicopter monitoring features can be the direct monitoring object, such as vibration signals, or the result after feature extraction of the monitoring object, such as the root mean square (RMS) of the vibration signal. However, the helicopter operating condition flight parameters and health monitoring features should be consistent in time.

[0039] Specifically, since the dimensions and orders of magnitude of the flight parameters under different operating conditions are different, they will have a certain impact on the mapping results of the monitoring features. Therefore, it is necessary to normalize the data so that the flight parameters under different operating conditions are on the same order of magnitude. The difference between the generated value and the true value of the monitoring feature is the residual. The monitoring feature mapping network takes minimizing the residual as its training objective.

[0040] Furthermore, multiple single-condition monitoring feature mapping networks will be obtained through the technical means provided by S200.

[0041] S300: For the trained single-condition monitoring feature mapping network, input the health samples under the corresponding conditions and calculate the health sample residual X. e0 Using the concept of statistical distribution, calculate X. e0 mean μ e and variance σ e ; Residual X e0 mean μ e and variance σ e The value should be approximately 0, representing the gap between the features generated by the feature mapping network for health sample monitoring and the real features.

[0042] S400: For the feature sample X = [x1, x2, ..., x] to be monitored for anomalies i [,...], Select the appropriate window length w L The monitored features are segmented using a sliding time window with a step size Δw, where w is the window length. L The number of internal sample points is n, and each time window has segmentation monitoring features X. r =[x i+1 ,x i+2 ,…,xi+n The length w of the sliding time window L The window length wL and step size Δw should be determined according to the monitoring requirements. If the timeliness of anomaly monitoring is required and the monitoring characteristics fluctuate greatly, the window length wL and step size Δw should be appropriately reduced. If the speed of anomaly monitoring is required and the monitoring characteristics are smooth, the window length wL and step size Δw should be appropriately increased.

[0043] Segmentation monitoring feature X under a specific time window r =[x i+1 ,x i+2 ,…,x i+n The operating condition parameters at the corresponding time are input into the operating condition clustering model for operating condition identification to determine the operating condition within that time window. Matching and mapping with the corresponding monitoring feature mapping network are then performed, and the generated monitoring feature X is obtained using the mapping network. g =[x i+1 ',x i+2 ',…,x i+n '], corresponding residual X e =X r -X g =[e i+1 ,e i+2 ,…,e i+n ], combining the ideas of statistical distribution, calculate X g mean μ g and variance σ g ;

[0044] Segmentation monitoring feature X under each time window r =[x i+1 ,x i+2 ,…,x i+n If it is a healthy sample, X r and generate monitoring feature X g It should have high consistency, and the corresponding residual X e =X r -X g The mean obtained in S3 should be μ. e The variance is σ e Statistical distribution.

[0045] S500: Based on segmentation monitoring feature X r Generate monitoring feature X g Residual X e and healthy sample residuals X e0 The relationship between X e =X r -X g X under conditions where the sample has no abnormalities e and X e0 They have the same distribution, that is, the mean is μ.e The variance is σ e It follows a normal distribution.

[0046] Segmentation and monitoring of feature X r Calculation of the upper control line (UCL), center line (CL), and lower control line (LCL) of the control limits. Given X... r =X g +X e Referring to the "3σ" criterion, X can be calculated. r The control limits are:

[0047]

[0048] Where μ r0 =E(X) g +X e )=f(μ g ,μ e ), Specifically:

[0049] μ r0 =f(μ g ,μ e )=μ g +μ e

[0050]

[0051] Wherein, UCL is the segmentation monitoring feature X r The upper control limit, CL is the segmentation monitoring feature X. r The center line, LCL is the segmentation monitoring feature X r The lower control limit. If segmented monitoring feature X r If the sample is healthy and there are no abnormalities, then X r =[x i+1 ,x i+2 ,…,x i+n It should fluctuate between LCL and UCL; correspondingly, if X r =[x i+1 ,x i+2 ,…,x i+n If any monitored feature exceeds the upper or lower control limit, it is determined that an anomaly has occurred within that time window.

[0052] S600: As the sliding time window moves, the control limits within the window will change synchronously with the different real-time operating conditions. This allows for dynamic monitoring of helicopter anomalies. This change will be step-like, meaning that within a specific window, the control limit will be a constant value.

[0053] In this embodiment, it should be noted that the above steps are implemented through computer programming, thereby enabling dynamic monitoring of helicopter anomalies based on operating conditions.

[0054] The technical solution conceived in this invention has the following advantages compared with the prior art:

[0055] This invention enables the identification of helicopter flight conditions and, based on this identification, performs anomaly monitoring. Helicopter health monitoring features can be directly monitored objects or results obtained after feature extraction from the monitored objects. Control limits consider the fluctuation range of health features and the error of the mapping network, resulting in more reasonable interval settings. By setting a sliding time window, real-time condition identification and corresponding control limit setting are achieved, improving monitoring accuracy. Those skilled in the art only need to apply this method to helicopter flight parameters and time-consistent health monitoring features to achieve dynamic anomaly monitoring of helicopters based on flight conditions.

[0056] While this application provides the method operation steps as described in the embodiments or flowcharts, more or fewer operation steps may be included based on conventional or non-inventive means. The order of steps listed in the embodiments is merely one possible order of execution among many steps and does not represent the only possible order. In actual device or client product execution, the methods shown in the embodiments or drawings may be executed sequentially or in parallel (e.g., in a parallel processor or multi-threaded processing environment, or even a distributed data processing environment). The terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, product, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, product, or apparatus. Without further limitations, the presence of other identical or equivalent elements in a process, method, product, or apparatus that includes said elements is not excluded.

[0057] The above embodiments are only for illustrating the technical concept and features of the present invention, and are intended to enable those skilled in the art to understand the content of the present invention and implement it accordingly. They should not be construed as limiting the scope of protection of the present invention. All equivalent transformations or modifications made in accordance with the spirit and essence of the present invention should be covered within the scope of protection of the present invention. Technical aspects, shapes, and structures not described in detail in this invention are all well-known technologies.

Claims

1. A method for dynamic monitoring of helicopter anomalies based on operating conditions, characterized in that, Includes the following steps: Select a sample set of flight parameters; Train the working condition clustering model based on the sample set; Training of the monitoring feature mapping network for single-condition sample sets is performed until the difference between the generated monitoring feature value and the true value is minimized. For the trained single-condition sample set monitoring feature mapping network, input healthy samples under the corresponding conditions, and calculate the residuals of the healthy samples. ,as well as mean and variance ; Perform time segmentation of the monitoring features for the abnormal sample X to obtain the segmented monitoring features under each time window. ; Input the operating condition parameters of the abnormal sample X at the corresponding time into the operating condition clustering model to obtain the monitoring characteristics of the abnormal sample. Corresponding residuals and monitoring characteristics of abnormal samples mean and variance ; Based on segmentation monitoring characteristics Generate monitoring features residual and healthy sample residuals The relationship between them, and the calculation of segmentation monitoring features The upper control limit, center line, and lower control limit are determined according to the following formula: ； in, and All are intermediate quantities. To segment monitoring features The upper control limit, To segment monitoring features The centerline, To segment monitoring features The lower control limit; Monitoring segmentation monitoring features The positional relationship between the upper control limit, center line, and lower control limit, when segmenting monitoring features. If the upper or lower control limit is exceeded, the current time window is considered to be abnormal.

2. The method for dynamic monitoring of helicopter anomalies based on operating conditions according to claim 1, characterized in that, The sample set of flight parameters includes at least: The training includes maintenance manuals, system structure and principles, expert experience, a flight parameter sample set with selected key operating condition flight parameters as input and corresponding operating condition labels as output, and a flight operating condition clustering model trained based on a Gaussian mixture model.

3. The method for dynamic monitoring of helicopter anomalies based on operating conditions according to claim 1, characterized in that, The training of the single-condition sample set monitoring feature mapping network takes the condition flight parameters as input and the corresponding health monitoring features as output. The health monitoring features are those of the directly monitored object or features of the monitored object after feature extraction. The flight parameters and health monitoring characteristics under the specified conditions are consistent in time.

4. The method for dynamic monitoring of helicopter anomalies based on operating conditions according to claim 1, characterized in that, The time segmentation for monitoring the features of abnormal sample X specifically involves selecting a certain window length. and step length The monitoring features are segmented using a sliding time window, so that each time window has segmented monitoring features. .

5. The method for dynamic monitoring of helicopter anomalies based on operating conditions according to claim 4, characterized in that, Window length of the sliding time window and step length The length of the sliding time window is inversely proportional to the timeliness requirement of anomaly monitoring. and step length It is inversely proportional to the fluctuation of the monitoring characteristics.

Images