A helicopter monitoring method and system based on mixed characteristic sound

Abstract

The application discloses a kind of based on mixed feature sound's helicopter monitoring method and system, it is related to helicopter application health and use monitoring technical field, solve the coupling degree of existing monitoring system and fuselage structure deep, cannot adapt to the problem of different machine type, its technical scheme main point is: system includes sensor subsystem and processing host, sensor subsystem is connected with processing host;Sensor subsystem includes acoustic array sensor, acoustic node sensor and exogenous sensor, acoustic array sensor is deployed in cabin, for collecting acoustic data in cabin, acoustic node sensor is deployed in specific position of cabin, for collecting acoustic data in specific position, exogenous sensor is connected with helicopter operating system, for collecting the operation data of helicopter;Processing host outputs monitoring information containing early warning information, abnormal sound source positioning information, fault type and the probability of the fault type;Using the detection network of containing acoustic sensor cooperates targeted database, can adapt to different machine type.

Description

Technical Field

[0001] This invention relates to the field of helicopter application health and usage monitoring technology, and more specifically, to a helicopter monitoring method and system based on mixed characteristic sound. Background Technology

[0002] With the increasing use of helicopters in military operations, disaster relief, and other fields, Helicopter Health and Usage Monitoring Systems (HUMS) are gaining importance. As an auxiliary system that spans the entire lifecycle of helicopter equipment, HUMS helps to predict malfunctions, identify potential problems, and monitor equipment status. The use of HUMS is of significant value in reducing the harm caused by malfunctions, improving maintenance and support capabilities, and ensuring crew safety. As a complex equipment system, the technologies applied in HUMS also offer valuable application examples for monitoring systems in other equipment.

[0003] In the current HUMS, the most important sensor is the contact vibration sensor, which measures the vibration signals of components such as the engine, transmission system, and rotor, and combines this with the helicopter's operating status, such as speed, engine speed, temperature and other parameters, to monitor the status of the entire helicopter system.

[0004] Existing vibration sensors, based on the principle of measuring solid vibration and acceleration signals, mostly employ fixed contact installations. This results in a deep coupling between the entire system and the fuselage structure. If the system needs to be installed on different helicopters, specific hardware adjustments are required. Furthermore, nodal-type accelerometers are relatively inflexible in installation; a small number offer limited fault location capabilities, while a large number lead to complex deployment. Mechanical waves propagate quickly in solids, have long wavelengths, and their propagation paths are limited and influenced by the mechanical structure, further impacting fault location capabilities and limiting analysis to fault mode analysis. Consequently, the overall system structure lacks flexibility, with limited room for further adjustments or expansions.

[0005] In view of this, the inventors propose a helicopter monitoring method and system based on mixed characteristic sound. Summary of the Invention

[0006] The purpose of this application is to provide a helicopter monitoring method and system based on hybrid characteristic sound. It uses acoustic sensors to establish a detection network covering the aircraft body to realize fault early warning, location and diagnosis, replacing the traditional fault detection scheme based on vibration signal, and can be widely adapted to different aircraft models.

[0007] This application first provides a helicopter monitoring method based on mixed characteristic sound. The above technical objective is achieved through the following technical solution: the method includes...

[0008] Acquire acoustic, operational, and structural data of the helicopter;

[0009] The acoustic data is preprocessed, at least one low-order feature data is extracted from the processed acoustic data, and the low-order feature data is compared with normal data in the database. If there is an anomaly, an early warning message is generated.

[0010] When the acoustic data is abnormal, the location information of the abnormal sound source is obtained by using the acoustic data, operational data, structural data, and the prior probability of the fault distribution in the cabin.

[0011] The acoustic data is filtered and enhanced using the abnormal sound source localization information to obtain acoustic feature data to be identified. The acoustic feature data to be identified and the running data are then input into a diagnostic model based on an attention mechanism to obtain the fault type and the probability of that fault type.

[0012] Monitoring results are generated based on the warning information, abnormal sound source location information, fault type, and probability of that fault type.

[0013] By adopting the above technical solution, comprehensive monitoring of fault early warning, fault location and fault diagnosis can be achieved based on acoustic data. Acoustic sensors can collect data without being coupled to the fuselage. The method can be adapted to different aircraft models, and the coverage of acoustic detection is larger, making it easier to build a detection network for the whole aircraft. Fault location is achieved by combining acoustic data, operational data, structural data and the prior probability of fault distribution in the cabin, which reduces the dependence on sensors, reduces the impact of propagation path structure and makes the location more accurate.

[0014] Furthermore, by utilizing the acoustic data, operational data, structural data, and the prior probability of fault distribution in the cabin, abnormal sound source localization information is obtained, including:

[0015] The spatial distribution probability of the sound source is obtained through the acoustic data and operational data. The historical fault location distribution probability is obtained through the structural data and the prior probability of the fault distribution in the cabin. The abnormal sound source location information is obtained through the spatial distribution probability of the sound source and the historical fault location distribution probability.

[0016] Furthermore, the acoustic data is filtered and enhanced using the abnormal sound source localization information to obtain acoustic feature data to be identified, including:

[0017] Based on the abnormal sound source localization information, acoustic data from the vicinity of the abnormal sound source location are filtered, and the acoustic data from the vicinity of the abnormal sound source location are enhanced to obtain acoustic feature data to be identified.

[0018] Furthermore, the acoustic feature data to be identified and the operational data are input into an attention-based diagnostic model to obtain the fault type and the probability of that fault type, which is derived through the following steps:

[0019] The acoustic feature data to be identified and the running data are converted into vectors of the same dimension;

[0020] An attention model is used to extract feature values ​​from the vectors of the same dimension, and weights are assigned to the feature values.

[0021] The feature values, including weights, are input into the diagnostic model to obtain the classification results: the fault type and the probability of that fault type.

[0022] Furthermore, it also includes updating the database with normal data through the acoustic data and operational data, and optimizing the attention-based diagnostic model through the acoustic feature data to be identified, operational data, and actual fault types.

[0023] Another aspect of this application provides a helicopter monitoring system based on mixed characteristic sound, the system including...

[0024] A sensor subsystem and a processing host, wherein the sensor subsystem is connected to the processing host;

[0025] The sensor subsystem includes an acoustic array sensor, an acoustic node sensor, and an external sensor. The acoustic array sensor is deployed inside the cabin to collect acoustic data within the cabin. The acoustic node sensor is deployed at a specific location in the cabin to collect acoustic data at that specific location. The external sensor is connected to the helicopter operating system to collect helicopter operating data.

[0026] The built-in program of the processing host is used to implement a helicopter monitoring method based on mixed characteristic sound as described above, and outputs monitoring information including early warning information, abnormal sound source location information, fault type and probability of the fault type.

[0027] Furthermore, the processing host includes: an acoustic signal preprocessing module, an abnormal signal rapid detection module, an abnormal sound source localization module, a fault diagnosis module, and a data maintenance and update module;

[0028] The acoustic signal preprocessing module is used to preprocess the acoustic data;

[0029] The abnormal signal rapid detection module is used to extract at least one low-order feature data from the processed acoustic data, compare the low-order feature data with normal data in the database, and generate early warning information if an abnormality is found.

[0030] The abnormal sound source localization module is used to obtain abnormal sound source localization information by means of the acoustic data, operational data, structural data and the prior probability of the fault distribution in the cabin when the acoustic data is abnormal.

[0031] The fault diagnosis module is used to filter and enhance the acoustic data through the abnormal sound source localization information to obtain acoustic feature data to be identified, and input the acoustic feature data to be identified and the running data into the attention mechanism-based diagnostic model to obtain the fault type and the probability of the fault type.

[0032] The data maintenance and update module is used to update the normal data of the database through the acoustic data and operational data, and to optimize the attention-based diagnostic model through the acoustic feature data to be identified, operational data, and actual fault types.

[0033] Furthermore, the abnormal sound source localization module includes:

[0034] The sound source spatial distribution probability calculation module is used to obtain the sound source spatial distribution probability through the acoustic data and operational data;

[0035] The historical fault location distribution probability calculation module is used to obtain the historical fault location distribution probability through the structural data and the prior probability of fault distribution in the cabin.

[0036] An abnormal sound source localization module is used to obtain abnormal sound source localization information through the spatial distribution probability of the sound source and the historical fault location distribution probability.

[0037] Furthermore, the fault diagnosis module includes:

[0038] The data filtering and enhancement module is used to filter acoustic data from the vicinity of the abnormal sound source location based on the abnormal sound source location information, and enhance the acoustic data in the direction of the abnormal sound source location to obtain acoustic feature data to be identified.

[0039] Furthermore, the fault diagnosis module includes:

[0040] The data filtering and enhancement module is used to filter acoustic data from the vicinity of the abnormal sound source location based on the abnormal sound source location information, and enhance the acoustic data in the direction of the abnormal sound source location to obtain acoustic feature data to be identified.

[0041] Compared with existing technologies, the helicopter monitoring method and system based on hybrid characteristic sound proposed in this application have the following advantages:

[0042] Beneficial effects:

[0043] This application achieves comprehensive detection of faults, including early warning, fault location, and fault diagnosis, based on acoustic data. The acoustic sensor can acquire acoustic data in the cabin in a non-contact manner, adapting to different aircraft models. For fault warning, low-order feature data is used to improve the speed of warning. For fault location, acoustic data, operational data, structural data, and the prior probability of fault distribution in the cabin are used to jointly locate abnormal sound sources, reducing reliance on sensor data and making location more reliable. For fault judgment, abnormal sound source location information is used for reverse filtering and enhancement of input information, combined with an attention-based diagnostic model to improve diagnostic accuracy and efficiency. This application also provides real-time updated steady-state data through online and offline database updates, and optimizes the attention-based diagnostic model based on the acoustic feature data to be identified, operational data, and actual fault types. Attached Figure Description

[0044] The accompanying drawings, which are included to provide a further understanding of embodiments of the invention and form part of this application, do not constitute a limitation thereof. In the drawings:

[0045] Figure 1 This is a schematic flowchart of a monitoring method provided in an embodiment of the present invention;

[0046] Figure 2 This is a schematic diagram of the structure of a detection system provided in an embodiment of the present invention;

[0047] Figure 3 This is a schematic diagram of an acoustic array sensor provided in an embodiment of the present invention;

[0048] Figure 4 This is a data processing flowchart of an acoustic array sensor provided in an embodiment of the present invention;

[0049] Figure 5 This is a data processing flowchart of an acoustic node sensor provided in an embodiment of the present invention;

[0050] Figure 6 This is a schematic diagram of the frequency characteristics of a helicopter operating noise according to an embodiment of the present invention;

[0051] Figure 7 This is a schematic diagram of the beamforming spatial spectrum of an acoustic array sensor detecting a moving target according to an embodiment of the present invention;

[0052] Figure 8 An abnormal sound source localization map obtained by combining acoustic data, operational data, structural data, and the prior probability of fault distribution in the cabin, as provided in an embodiment of the present invention;

[0053] Figure 9 This is a flowchart of obtaining abnormal sound source localization information provided in an embodiment of the present invention;

[0054] Figure 10 A schematic diagram of the structure of an attention-based diagnostic model provided in an embodiment of the present invention. Figure 1 ;

[0055] Figure 11 A schematic diagram of the structure of an attention-based diagnostic model provided in an embodiment of the present invention. Figure 2 ;

[0056] Figure 12 This is a task allocation diagram of a monitoring system provided in an embodiment of the present invention. Detailed Implementation

[0057] In the following, the terms “comprising” or “may include” as used in the various embodiments of this application indicate the presence of the claimed function, operation, or element, and do not limit the addition of one or more functions, operations, or elements. Furthermore, as used in the various embodiments of this application, the terms “comprising,” “having,” and their cognates are intended only to indicate a specific feature, number, step, operation, element, component, or combination of the foregoing, and should not be construed as primarily excluding the presence of one or more other features, numbers, steps, operations, elements, components, or combinations of the foregoing, or the possibility of adding one or more combinations of the foregoing.

[0058] In various embodiments of this application, the expression "or" or "at least one of B and / or C" includes any combination or all combinations of the words listed simultaneously. For example, the expression "B or C" or "at least one of B and / or C" may include B, may include C, or may include both B and C.

[0059] It should be noted that if a description refers to "connecting" a component to another component or "connecting" it to another component, then the first component can be directly connected to the second component, and a third component can be "connected" between the first and second components. Conversely, when a component is "directly connected" to another component or "directly connected" to another component, it can be understood that there is no third component between the first and second components.

[0060] The terminology used in the various embodiments of this application is for the purpose of describing particular embodiments only and is not intended to limit the various embodiments of this application. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Unless otherwise defined, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by one of ordinary skill in the art to which the various embodiments of this application pertain. The terms (such as those defined in a generally used dictionary) are to be interpreted as having the same meaning as in the context of the relevant technical field and are not to be interpreted as having an idealized or overly formal meaning, unless clearly defined in the various embodiments of this application.

[0061] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the embodiments and accompanying drawings. The illustrative embodiments and descriptions of this application are only for explaining this application and are not intended to limit this application.

[0062] Traditional helicopter fault detection commonly uses vibration or acceleration sensors. These sensors are fixed contact type and deeply coupled with the fuselage structure. The problems are: 1. Fixed contact type installation has poor universality and requires adjustment when changing aircraft models; 2. Fixed contact sensor layout is complex and it is difficult to form a comprehensive detection network; 3. The sensor is coupled to the fuselage and is easily affected by the structure during detection, resulting in poor fault location capability.

[0063] Acoustic and vibration signals both originate from mechanical waves, and the vibration of equipment will also propagate through the air, which can be measured by acoustic sensors. Furthermore, the applicant discovered that when a helicopter accident occurs, failures in transmission components such as the engine, rotor, and tail rotor are the most impactful types of failures on the helicopter, and the acoustic data generated by these components are very distinctive.

[0064] Therefore, based on the defects in the existing technology and the acoustic characteristics of helicopter faults, the inventors propose a helicopter monitoring method and system based on mixed characteristic sound. The aim is to provide helicopters with comprehensive, accurate and flexible condition monitoring and life cycle management through a combination of non-contact acoustic monitoring and neural networks, so as to realize the functions of fault early warning, fault location and fault diagnosis.

[0065] Example 1:

[0066] This embodiment provides a helicopter monitoring method based on hybrid characteristic sound. By introducing non-contact acoustic monitoring, it achieves fault early warning, fault location, and fault diagnosis functions. The method includes:

[0067] S1. Acquire acoustic, operational, and structural data of the helicopter;

[0068] S2. Preprocess the acoustic data, extract at least one low-order feature data from the processed acoustic data, compare the low-order feature data with normal data in the database, and generate early warning information if there is an anomaly.

[0069] S3. When the acoustic data is abnormal, obtain the abnormal sound source location information by using the acoustic data, operational data, structural data and the prior probability of the fault distribution in the cabin;

[0070] S4. The acoustic data is filtered and enhanced using the abnormal sound source localization information to obtain acoustic feature data to be identified. The acoustic feature data to be identified and the running data are input into the attention mechanism-based diagnostic model to obtain the fault type and the probability of the fault type.

[0071] S5. Generate monitoring results based on the warning information, abnormal sound source location information, fault type and probability of the fault type.

[0072] Specifically, see Figure 1 As shown, in step S1, an acoustic sensor network covering the fuselage is constructed using a non-fixed contact method to collect acoustic data from multiple directions and points. Specifically, the acoustic sensor can be an acoustic array sensor or an acoustic node sensor. These sensors can collect acoustic data without coupling to the fuselage. If necessary, multiple acoustic node sensors can be arranged in an array to replace the acoustic array sensor. Operational data is directly acquired from the helicopter system through external sensors. This operational data includes, but is not limited to, the helicopter's flight status, steady-state signals, speed, altitude, and flight attitude. Structural data can come from the aircraft's structural design drawings, containing the aircraft's spatial structural features.

[0073] After obtaining the aforementioned acoustic data, operational data, and structural data, low-order features are extracted from the massive amount of data to provide rapid early warning. Specifically, see step S2: preprocess the acoustic data, extract at least one low-order feature data from the processed acoustic data, compare the low-order feature data with normal data in the database, and generate early warning information if an anomaly is found.

[0074] Specifically, in step S2, the acoustic data preprocessing mainly involves suppressing the steady-state noise of the helicopter and extracting weak measurement signals. The steady-state noise of the helicopter is related to its operating state, exhibits significant frequency domain characteristics, and can affect the detection results. Please refer to [link to relevant documentation]. Figure 6 As shown, Figure 6This diagram illustrates the frequency response of helicopter operating noise, showing the presence of harmonics. In practical scenarios, common frequency domain filtering methods such as comb filtering and spectral subtraction can be used for noise suppression. Furthermore, when using array sensors for signal reception, spatial filtering can be performed in the spatial (azimuth) domain in addition to frequency domain filtering. The signal characteristic information required for noise suppression can be provided by databases and flight status and steady-state signals from operational data.

[0075] After the above preprocessing, the steady-state noise of the acoustic data is suppressed, and the weak measurement signal is extracted as a useful signal. Although the extraction of the weak measurement signal through preprocessing reduces the content of the acoustic data, it is difficult to achieve rapid early warning because the acoustic data contains many features. Therefore, the inventors proposed a rapid early warning system based on low-order feature data extraction. The purpose is to compare some low-order features (such as energy, spectrum, time, etc.) that can be quickly extracted from the signal with the steady-state features of the signal before performing complex signal processing, so as to provide the system with a rapid response early warning function.

[0076] For example, in specific implementation scenarios, power spectrum features of acoustic data can be extracted to construct short-time power spectrum statistical features for a specified set of frequencies of interest:

[0077] F(t) = ZA(f)S(t, f) for f ∈ F *

[0078] Where F(t) is the statistical characteristic of the power spectrum at time t; F * It is the set of frequency bands that the feature is interested in. Other frequency bands do not need to be included in the calculation. A(f) is the weighting of these frequencies by the feature. Different dimensional features F(t) can be constructed by designing the dimensions of A(f) as needed. s(t,f) is the short-time Fourier transform of the preprocessed signal at time t.

[0079] When the above characteristics undergo a significant change in a short period of time, an early warning message is generated immediately to issue an alert, without waiting for the subsequent processes to run.

[0080] It should be noted that in other specific implementation scenarios, methods such as energy monitoring, power spectrum matching, and harmonic set feature analysis can also be used for rapid detection. Furthermore, the low-order features obtained in this stage can be incorporated into the fault diagnosis process for added convenience.

[0081] After achieving rapid early warning through the above steps, this method can further achieve fault location through step S3. Step S3: When the acoustic data is abnormal, the abnormal sound source location information is obtained through the acoustic data, operational data, structural data, and the prior probability of the fault distribution in the cabin. Step S3 further includes:

[0082] The spatial distribution probability of the sound source is obtained through the acoustic data and operational data. The historical fault location distribution probability is obtained through the structural data and the prior probability of the fault distribution in the cabin. The abnormal sound source location information is obtained through the spatial distribution probability of the sound source and the historical fault location distribution probability.

[0083] Specifically, the main task of step S3 is to estimate the location of the anomaly. First, the spatial distribution probability P(p) of the sound source is obtained using acoustic and operational data: the acoustic node sensor estimates the region of the anomalous sound source based on the anomalous signal intensity, the acoustic array sensor estimates the azimuth angle of the anomalous sound source, and through array processing techniques based on beamforming, MUSIC, and CLEAN algorithms, the distribution characteristics of the anomalous sound source within the helicopter space are given in the form of a spatial spectrum, such as... Figure 7 As shown, this can be further reflected in the spatial probability distribution. Figure 7 This is a schematic diagram of the beamforming spatial spectrum when an acoustic array sensor detects a moving target. Figure 3 The spatial spectrum results of a planar acoustic array sensor on the left detecting a moving target in a corridor-shaped space of approximately 3m × 3m.

[0084] Secondly, the historical fault location distribution probability P(p|M) is obtained through the structural data and the prior probability of the fault distribution in the cabin: for a specific fault type, it will only appear in a specific location in the helicopter cabin (or is more likely to appear). This prior information can also be represented in the form of a distribution probability.

[0085] In summary, this example employs a joint probability density-based localization method. It obtains the abnormal sound source localization information P(p, M) by jointly analyzing the spatial distribution probability P(p) of the sound source and the historical fault location distribution probability P(p|M). Specifically, the probability distribution of a certain fault M within the cabin can be expressed as P(p, M) = P(p)P(p|M), where P(p) is the probability distribution density of position p obtained from array measurements, and P(p|M) is the prior probability of fault M at position p within the cabin. The results are as follows: Figure 8 As shown in the diagram on the left, the beam represents the measurement probability distribution, the frame structure represents the probability range of possible fault occurrences, and the combined scatter plot represents the fault probability distribution. The larger the area of ​​a scatter plot, the greater the probability of a fault occurring at that point. Scatter plots with an area greater than a certain threshold can be used as abnormal sound source location information.

[0086] After obtaining the abnormal sound source localization information through the aforementioned joint probability density, the fault type needs to be further determined through step S4. Step S4: The acoustic data is filtered and enhanced using the abnormal sound source localization information to obtain acoustic feature data to be identified. This acoustic feature data and the operational data are then input into an attention-based diagnostic model to obtain the fault type and its probability. This includes:

[0087] S41. Filtering and enhancing the acoustic data using abnormal sound source location information to obtain acoustic feature data to be identified includes: filtering acoustic data from the vicinity of the abnormal sound source location based on the abnormal sound source location information, and enhancing the acoustic data from the vicinity of the abnormal sound source location to obtain acoustic feature data to be identified.

[0088] Specifically, abnormal sound source localization information can guide the screening and enhancement of input data (acoustic feature data to be identified) used to determine the type of fault, improving both efficiency and accuracy. First, acoustic data near the abnormal sound source location is screened from a large pool of acoustic data. For example, if the abnormal sound source location is confirmed to be point A in the cabin, then selecting acoustic data from point A and its vicinity as input is more representative. Simultaneously, the acoustic data from point A and its vicinity can be enhanced in the direction of A to further strengthen the data features. When using an acoustic array sensor, beamforming methods can be used to enhance the acoustic data in the direction of A.

[0089] S42. Input the acoustic feature data to be identified and the operational data into the attention-based diagnostic model to obtain the fault type and the probability of that fault type, which is obtained through the following steps:

[0090] The acoustic feature data to be identified and the running data are converted into vectors of the same dimension;

[0091] An attention model is used to extract feature values ​​from the vectors of the same dimension, and weights are assigned to the feature values.

[0092] The feature values, including weights, are input into the diagnostic model to obtain the classification results: the fault type and the probability of that fault type.

[0093] Specifically, the fault diagnosis in this example is based on neural networks and heterogeneous data. For example... Figure 9 As shown, step S42 involves fusing the acoustic feature data to be identified and the operational data. The acoustic feature data itself comes from different acoustic sensors and also needs to be fused. Therefore, the heterogeneous input data is first processed, transforming it into a vector of the same dimension for fault diagnosis. Figure 2 Taking the schematic diagram of the sensing system shown below as an example, the detailed steps are as follows:

[0094] 1) Define the network structure based on the helicopter's structure, sensor composition, and fault types. Figure 2 The sensor distribution shown is mainly divided into three types. For ease of explanation, the number of sensors is defined as three: one acoustic array sensor, one acoustic node sensor, and one external sensor.

[0095] 2) Based on the sparse distribution method, the information received by the acoustic array sensor, acoustic node sensor and external sensor on the helicopter is converted into vectors of the same dimension.

[0096] Since high-precision location information calculation takes time, and the signal sampling rate is 10kHz, simple sequence splicing can easily lose some valuable information. Therefore, preprocessing of both information is necessary. The location information is linearly transformed into a 1024-dimensional vector. Specifically, the n sets of acquired location information are multiplied by a matrix of [n, 1024] to achieve information sparsity, where the parameters in the matrix are within the interval [n, 1024]. The random number in the image; the sound signal is sliced ​​to obtain 1024 sampling points;

[0097] The acquired array sensor position information P 1i Measurement data information R 1i Node sensor location information P 2i Measurement data information R 2i External sensor location information P 3i Measurement data information R 3i Convert to in It is a vector with a dimension of 1024 after preprocessing (encoding or slicing). It should be noted that the position information of the above sensors comes from prior information entered into the database in advance.

[0098] 3) Normalize the vector to between 0 and 1.

[0099] After the acoustic feature data and operational data to be identified are converted into vectors of the same dimension through the above process, feature values ​​need to be extracted from numerous vectors and weighted according to their importance. This allows the diagnostic model to focus on the data that has a greater impact on the results and reduces the amount of computation. Therefore, this example uses a diagnostic model based on the attention mechanism.

[0100] 4) Define the attention model and assign weights to each feature. Detailed steps are as follows:

[0101] a) The attention model includes a feature extraction module and a feature attention module. First, the sparsely processed information is fed into the feature extraction module, which contains one convolutional layer, two convolutional blocks, and one average pooling layer. Each convolutional block contains three convolutional layers, a batch normalization layer, and an activation layer, such as... Figure 10 As shown;

[0102] b) The extracted features are fed into the attention model. The information selection network consists of convolutional modules and a softmax layer, as shown below. Figure 11 As shown, Softmax is used to convert the input features into numbers from 0 to 1 as the weights of the features. The weights are connected to form a feature selection matrix and multiplied with the features to achieve the selection of features.

[0103] 5) Define the layers and parameters in the diagnostic model, and generate training samples for each category to train the diagnostic model. The construction process of the diagnostic model is as follows:

[0104] The diagnostic model consists of two hidden layers and two fully connected layers. Each hidden layer comprises three parts: a convolutional layer, a ReLU activation function, and a max-pooling layer. The two convolutional layers contain 64 and 128 kernels respectively, with a size of [3×3] and a stride of 2. The first fully connected layer uses ReLU activation and has a total of 2048 output neurons. The second fully connected layer uses Softmax activation to compute the posterior probability for each class. Adam gradient is used as the optimizer during training, with cross-entropy as the cost function.

[0105] 6) Convert the actual acoustic data and operational data measured by the sensors on the helicopter into vectors of dimension 1024 and normalize them. Then, put them into the trained attention-based diagnostic model to output the fault type and the probability of that fault type.

[0106] Finally, step S5 involves generating monitoring results based on the warning information, abnormal sound source location information, fault type, and the probability of that fault type. Specifically, the monitoring results can be output and interacted with through the host's output structure, and the output and interaction methods can include interactive devices such as alarms, voice prompts, and displays.

[0107] In some possible embodiments, the method further includes: S6, updating the normal data in the database using the acoustic data and operational data, and optimizing the attention-based diagnostic model using the acoustic feature data to be identified, operational data, and actual fault types. Specifically:

[0108] Due to changes in flight conditions and the helicopter's own operational status, the helicopter's steady-state characteristics are not static. Therefore, the normal-state data in the database can be updated using acoustic and operational data, and the attention-based diagnostic model can be optimized using the acoustic feature data to be identified, operational data, and actual fault types. Database maintenance and updates include both online and offline modes.

[0109] a) Online updates aim to provide accurate routine data as a reference for signal preprocessing and fault diagnosis during flight. Online updates need to extract steady-state operational noise characteristics from data over a period of time, and also need to assist in extracting steady-state noise suppression based on changes in flight conditions.

[0110] b) Offline maintenance: After the helicopter returns from its mission, analyze the sensor input data and diagnostic results, and save the sensor input data and diagnostic results to the database to complete the database update; further, for cases of diagnostic errors, send the acoustic feature data to be identified, the operating data and the actual fault type to the attention-based diagnostic model to continue to optimize the model. When the model's recognition accuracy reaches the target, send the data to the database for storage to complete the model optimization and database update.

[0111] It should be noted that online and offline updates have different focuses. Online updates need to balance efficiency and accuracy, forming reliable steady-state characteristics in real time; while offline updates have less demanding requirements for efficiency, and can more deeply assess the aircraft's condition by comparing the differences between normal state characteristics and current flight characteristics, in an attempt to discover potential failure risks.

[0112] This example proposes a helicopter monitoring method based on hybrid feature sound, combining acoustic, operational, and structural data of the helicopter to achieve fault early warning, fault location, and fault diagnosis. The sensors acquiring acoustic data can be deployed non-fixed within the cabin, adapting to different aircraft models. This example uses low-order feature data for fault early warning, improving the speed of fault prediction. It uses a multi-dimensional, comprehensive approach to locate abnormal sound sources by jointly analyzing acoustic, operational, and structural data, along with the prior probability of fault distribution within the cabin, reducing reliance on sensor data and making the location more reliable. This example uses abnormal sound source location information to filter and enhance input information, and combines it with an attention-based diagnostic model for fault diagnosis, improving diagnostic accuracy and efficiency. This example provides real-time updated routine data through online and offline database updates, and optimizes the attention-based diagnostic model based on the acoustic feature data to be identified, operational data, and actual fault types.

[0113] Compared to existing vibration monitoring methods, this example uses acoustic signals for monitoring. The acoustic sensors do not need to be in fixed contact with the fuselage, making them adaptable to different aircraft models. A detection network covering the cabin and fuselage can be built, resulting in more comprehensive detection. Furthermore, the joint positioning of acoustic signals and multidimensional probability distribution is less affected by structure compared to vibration signals, leading to more accurate fault location.

[0114] Example 2

[0115] This embodiment provides a helicopter monitoring system based on mixed characteristic sound, building upon Embodiment 1, to implement the helicopter monitoring method based on mixed characteristic sound described in Embodiment 1. The system includes:

[0116] A sensor subsystem and a processing host, wherein the sensor subsystem is connected to the processing host;

[0117] The sensor subsystem includes an acoustic array sensor, an acoustic node sensor, and an external sensor. The acoustic array sensor is deployed inside the cabin to collect acoustic data within the cabin. The acoustic node sensor is deployed at a specific location in the cabin to collect acoustic data at that specific location. The external sensor is connected to the helicopter operating system to collect helicopter operating data.

[0118] The built-in program of the processing host is used to implement the helicopter monitoring method based on mixed characteristic sound as described in Example 1, and outputs monitoring information including early warning information, abnormal sound source location information, fault type and probability of the fault type.

[0119] Specifically, see Figure 2 As shown, Figure 2 This is a schematic diagram of the monitoring system. Figure 2 The diagram demonstrates two deployment methods for array sensors (wall-mounted and tabletop). In a real system, only one array sensor is needed, and other deployment methods and locations can be used. For clarity, the scale of each device in the diagram is larger than in the actual system.

[0120] First, the sensor subsystem of the detection system will be described. As the sensing component of the system, the sensor subsystem mainly consists of an acoustic array sensor, acoustic node sensors, and external sensors. The components are as follows:

[0121] 1) Acoustic array sensor

[0122] The acoustic array sensor, capable of direction finding of spatial sound sources and possessing strong acoustic signal acquisition capabilities, is the core measurement device of this system. Within the system, the acoustic array sensor is deployed within the cabin, and can be positioned on the bulkhead, roof, or floor as needed. When the cabin's load density is not too high, the direction finding results, combined with cabin structure and sensor location information, can provide localization results for abnormal sound sources. The structure of the acoustic array sensor is as follows... Figure 3 As shown in the figure, the left side is a planar acoustic array sensor with a central camera (wall-mounted type), and the right side is a stereo acoustic array sensor. In actual use, either a planar or stereo acoustic array sensor can be selected according to the needs.

[0123] 2) Acoustic node sensor

[0124] Acoustic node sensors, which can employ either accelerometers or acoustic sensors, are primarily deployed near engines, transmission systems, rotors, or other critical structures to collect acoustic signals from specific locations. The deployment location and number of sensors can be determined based on requirements. Compared to acoustic array sensors, node sensors have a more specific acquisition range and can also provide certain location information based on their deployment location.

[0125] 3) External sensors

[0126] External sensors refer to sensors that are different from the two types mentioned above. External sensors mainly collect measurement data from the aircraft's own systems as well as some data from external sources. These can include flight status (speed, altitude, flight attitude, etc.), engine temperature, and engine speed.

[0127] The acoustic array sensor and acoustic node sensor systems in the system will be connected to the processing host via wired lines, while optional external sensors will also be connected to the processing host through reserved interfaces. The acoustic data collected by the sensor systems will undergo different processing flows and be processed before being passed to subsequent steps or having features extracted and passed to subsequent steps, such as... Figure 4 and Figure 5 As shown.

[0128] The data aggregation and processing section of the system is the processing host. The processing host includes: an acoustic signal preprocessing module, an abnormal signal rapid detection module, an abnormal sound source localization module, a fault diagnosis module, and a data maintenance and update module;

[0129] The acoustic signal preprocessing module is used to preprocess the acoustic data;

[0130] The abnormal signal rapid detection module is used to extract at least one low-order feature data from the processed acoustic data, compare the low-order feature data with normal data in the database, and generate early warning information if an abnormality is found.

[0131] The abnormal sound source localization module is used to obtain abnormal sound source localization information by means of the acoustic data, operational data, structural data and the prior probability of the fault distribution in the cabin when the acoustic data is abnormal.

[0132] The fault diagnosis module is used to filter and enhance the acoustic data through the abnormal sound source localization information to obtain acoustic feature data to be identified, and input the acoustic feature data to be identified and the running data into the attention mechanism-based diagnostic model to obtain the fault type and the probability of the fault type.

[0133] The data maintenance and update module is used to update the normal data of the database through the acoustic data and operational data, and to optimize the attention-based diagnostic model through the acoustic feature data to be identified, operational data, and actual fault types.

[0134] Specifically, for the acoustic signal preprocessing module, its main task is to suppress the steady-state noise of the helicopter and extract weak measurement signals. Helicopters themselves generate significant noise, which can affect measurements. However, due to the operational characteristics of helicopters, their steady-state noise exhibits relatively obvious frequency domain characteristics, and these characteristics are correlated with the flight state, such as... Figure 6 As shown, helicopters exhibit harmonic characteristics. In practical scenarios, common frequency domain filtering methods, such as comb filtering and spectral subtraction, can be used for noise suppression. Furthermore, since array sensors are used for signal reception, spatial filtering can be performed in the spatial (azimuth) domain in addition to frequency domain filtering. The signal characteristic information required for noise suppression can be provided by databases and flight status and steady-state signals from operational data.

[0135] For the rapid anomaly signal detection module, its main task is to compare some easily extractable low-order features of the signal (such as energy, spectrum, time, etc.) with the signal's steady-state features before performing complex signal processing, providing the system with a rapid early warning function. For example, the power spectrum matching method can construct short-time power spectrum statistical features for a specified set of frequencies of interest.

[0136] F(t)=∑A(f)S(t,f)for f∈F *

[0137] Where F(t) is the power spectrum statistical characteristic at time t; F * It is the set of frequency bands that the feature is interested in. Other frequency bands do not need to be included in the calculation. A(f) is the weighting of these frequencies by the feature. Different dimensional features F(t) can be constructed by designing the dimensions of A(f) as needed. S(t,f) is the short-time Fourier transform of the preprocessed signal at time t.

[0138] When the aforementioned characteristics undergo significant sudden changes within a short period, the system immediately issues an early warning without waiting for subsequent processes to run. This system can also employ methods such as energy monitoring, power spectrum matching, and harmonic set feature analysis for rapid early warning. If required by the algorithm, low-order features obtained in this stage can also be incorporated into the fault diagnosis process.

[0139] The abnormal sound source localization module includes:

[0140] The sound source spatial distribution probability calculation module is used to obtain the sound source spatial distribution probability through the acoustic data and operational data;

[0141] The historical fault location distribution probability calculation module is used to obtain the historical fault location distribution probability through the structural data and the prior probability of fault distribution in the cabin.

[0142] An abnormal sound source localization module is used to obtain abnormal sound source localization information through the spatial distribution probability of the sound source and the historical fault location distribution probability.

[0143] Specifically, the main task of this part is to estimate the location of the anomaly. First, the acoustic node sensors estimate the region of the anomalous sound source based on the intensity of the anomalous signal. The acoustic array sensors can estimate the azimuth angle of the anomalous sound source. Through array processing techniques based on algorithms such as beamforming, MUSIC, and CLEAN, the distribution characteristics of the anomalous sound source within the helicopter space are given in the form of a spatial spectrum, such as... Figure 7 As shown, Figure 2 The left-side planar acoustic array sensor detects a segment of the spatial spectrum of a moving target within a corridor-shaped space of approximately 3m × 3m. This spectrum can be further reflected in the spatial probability distribution. Secondly, for characteristic fault types, they only occur (or are more likely to occur) at specific locations within the helicopter cabin; this prior information can also be represented in the form of probability distributions. Therefore, in anomaly localization, this system adopts a localization method based on joint probability density: for example, the probability distribution of a certain fault M within the cabin can be expressed as P(p, M) = P(p)P(p|M), where P(p) is the probability distribution density of position p obtained from array measurements, and P(p|M) is the prior probability of fault M at position p within the cabin. The results are similar to... Figure 8 As shown in the diagram. Once the location result is determined, it can also provide feedback in subsequent steps to help determine the likelihood of a fault.

[0144] The fault diagnosis module includes:

[0145] The data filtering and enhancement module is used to filter acoustic data from the vicinity of the abnormal sound source location based on the abnormal sound source location information, and enhance the acoustic data in the direction of the abnormal sound source location to obtain acoustic feature data to be identified.

[0146] The heterogeneous feature fusion module is used to convert the acoustic feature data to be identified and the running data into vectors of the same dimension;

[0147] The attention mechanism filtering module is used to extract feature values ​​from the vectors of the same dimension using an attention model and assign weights to the feature values.

[0148] The fault identification module is used to input the weighted feature values ​​into the diagnostic model to obtain the classification result: the fault type and the probability of that fault type.

[0149] Specifically, by integrating existing information, the processing center performs in-depth fault diagnosis, such as... Figure 9 As shown. The fault identification module in the system adopts intelligent judgment technology based on neural networks. It mainly consists of the following steps:

[0150] Step 1) Define the network structure based on the helicopter structure, sensor system composition, and fault types. Figure 2 The sensor system shown is distributed in three main categories. For ease of explanation, the number of sensors is defined as three: one acoustic array sensor, one acoustic node sensor, and one external sensor.

[0151] Step 2) Convert the information received by the acoustic array sensor, acoustic node sensor and external sensor on the helicopter into vectors of the same dimension according to the sparse distribution method.

[0152] Since high-precision location information calculation takes time, and the signal sampling rate is 10kHz, simple sequence splicing can easily lose some valuable information. Therefore, preprocessing of both information is necessary. The location information is linearly transformed into a 1024-dimensional vector. Specifically, the n sets of acquired location information are multiplied by a matrix of [n, 1024] to achieve information sparsity, where the parameters in the matrix are within the interval [n, 1024]. The random number in the image; the sound signal is sliced ​​to obtain 1024 sampling points.

[0153] The acquired acoustic array sensor position information P 1i Measurement data information R 1i Sound node sensor location information P 2i Measurement data information R 2i External sensor location information P 3i Measurement data information R 3i Convert to in It is a vector with a dimension of 1024 after preprocessing (encoding or slicing). It should be noted that the position information of the above sensors comes from prior information entered into the database in advance.

[0154] Step 3) Normalize the vector to between 0 and 1.

[0155] Step 4) Define the attention model and complete the weight allocation for each feature.

[0156] a) The attention model includes a feature extraction module and a feature attention module. First, the sparsely processed information is fed into the feature extraction module, which contains one convolutional layer, two convolutional blocks, and one average pooling layer. Each convolutional block contains three convolutional layers, a batch normalization layer, and an activation layer, such as... Figure 10 As shown.

[0157] b) The extracted features are fed into the attention model. The information selection network consists of convolutional modules and a softmax layer, as shown below. Figure 11 As shown, Softmax is used to convert the input features into numbers from 0 to 1 as the weights of the features. The weights are connected to form a feature selection matrix and multiplied with the features to achieve the selection of features.

[0158] Step 5) Define the layers and parameters in the classification network, and generate training samples for each category to train the model.

[0159] The classification model consists of two hidden layers and two fully connected layers. Each hidden layer comprises three parts: a convolutional layer, a ReLU activation function, and a max-pooling layer. The two convolutional layers contain 64 and 128 kernels respectively, with a size of [3×3] and a stride of 2. The first fully connected layer uses ReLU activation and has a total of 2048 output neurons. The second fully connected layer uses Softmax activation to calculate the posterior probability for each class. Adam gradient is used as the optimizer during training, with cross-entropy as the cost function.

[0160] Step 6) Perform actual helicopter fault detection based on the trained model.

[0161] The acoustic data and location information actually measured by the sensors on the helicopter are converted into a vector of dimension 1024 in step 2 and normalized. Then, they are put into the model trained in steps 4 and 5 and classified according to the anomaly category with the highest probability in the output of the classification network.

[0162] Step 7) Based on the early warning information, abnormal sound source location information, fault type, and probability of that fault type, prepare the output results. After processing by the host, the monitoring results are output and interacted with through the host's output structure. In addition to data and signal output, human-machine interaction methods can include various interactive devices such as alarms, voice prompts, and displays. The application and task allocation of the entire system are as follows: Figure 12 As shown.

[0163] For the data maintenance and update module, the steady-state characteristics of the helicopter are not static due to changes in flight status and the helicopter's own operational status. Therefore, the normal data in the database can be updated using acoustic data and operational data, and the attention-based diagnostic model can be optimized using acoustic feature data to be identified, operational data, and actual fault types. Database maintenance and updates include both online and offline maintenance modes.

[0164] a) Online updates aim to provide accurate routine data as a reference for signal preprocessing and fault diagnosis during flight. Online updates need to extract steady-state operational noise characteristics from data over a period of time, and also need to provide assistance for noise characteristic extraction based on changes in flight conditions.

[0165] b) Offline maintenance: After the helicopter returns from its mission, analyze the sensor input data and diagnostic results, and save the sensor input data and diagnostic results to the database to complete the database update; further, for cases of diagnostic errors, send the acoustic feature data to be identified, the operating data and the actual fault type to the attention-based diagnostic model to continue to optimize the model. When the model's recognition accuracy reaches the target, send the data to the database for storage to complete the model optimization and database update.

[0166] In contrast, online updates and offline maintenance have different focuses. Online maintenance needs to balance efficiency and accuracy, providing reliable feature support in real time; while offline maintenance has less demanding requirements for efficiency and can more deeply assess the aircraft's condition by comparing the differences between normal state features and current flight features, aiming to discover potential failure risks.

[0167] Compared to vibration or acceleration sensors commonly used in traditional fault detection, acoustic detection has a wider detection range. Furthermore, the structure and operational characteristics of helicopters result in relatively consistent noise levels, which is beneficial for noise suppression and feature identification. This system combines non-contact acoustic monitoring with neural network methods to maximize hardware versatility while achieving diagnostic functionality, enabling adaptation to different helicopter models.

[0168] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A helicopter monitoring method based on hybrid characteristic sound, characterized by: include Acquire acoustic, operational, and structural data of the helicopter; The acoustic data is preprocessed, at least one low-order feature data is extracted from the processed acoustic data, and the low-order feature data is compared with normal data in the database. If there is an anomaly, an early warning message is generated. When the acoustic data is abnormal, the location information of the abnormal sound source is obtained by using the acoustic data, operational data, structural data, and the prior probability of the fault distribution in the cabin. The acoustic data is filtered and enhanced using the abnormal sound source localization information to obtain acoustic feature data to be identified. The acoustic feature data to be identified and the running data are then input into a diagnostic model based on an attention mechanism to obtain the fault type and the probability of that fault type. Monitoring results are generated based on the warning information, abnormal sound source location information, fault type, and probability of that fault type.

2. The helicopter monitoring method based on mixed characteristic sound according to claim 1, characterized in that: By utilizing the acoustic data, operational data, structural data, and prior probabilities of fault distribution in the cabin, abnormal sound source localization information is obtained, including: The spatial distribution probability of the sound source is obtained through the acoustic data and operational data. The historical fault location distribution probability is obtained through the structural data and the prior probability of the fault distribution in the cabin. The abnormal sound source location information is obtained through the spatial distribution probability of the sound source and the historical fault location distribution probability.

3. The helicopter monitoring method based on mixed characteristic sound according to claim 1, characterized in that: The acoustic data is filtered and enhanced using the abnormal sound source localization information to obtain acoustic feature data to be identified, including: Based on the abnormal sound source localization information, acoustic data from the vicinity of the abnormal sound source location are filtered, and the acoustic data from the vicinity of the abnormal sound source location are enhanced to obtain acoustic feature data to be identified.

4. The helicopter monitoring method based on mixed characteristic sound according to claim 3, characterized in that: The acoustic feature data to be identified and the operational data are input into an attention-based diagnostic model to obtain the fault type and the probability of that fault type, which is achieved through the following steps: The acoustic feature data to be identified and the running data are converted into vectors of the same dimension; An attention model is used to extract feature values ​​from the vectors of the same dimension, and weights are assigned to the feature values. The feature values, including weights, are input into the diagnostic model to obtain the classification results: the fault type and the probability of that fault type.

5. The helicopter monitoring method based on hybrid characteristic sound according to claim 1, characterized in that: It also includes updating the database with normal data through the acoustic data and operational data, and optimizing the attention-based diagnostic model through the acoustic feature data to be identified, operational data, and actual fault types.

6. A helicopter monitoring system based on hybrid characteristic sound, characterized by: include A sensor subsystem and a processing host, wherein the sensor subsystem is connected to the processing host; The sensor subsystem includes an acoustic array sensor, an acoustic node sensor, and an external sensor. The acoustic array sensor is deployed inside the cabin to collect acoustic data within the cabin. The acoustic node sensor is deployed at a specific location in the cabin to collect acoustic data at that specific location. The external sensor is connected to the helicopter operating system to collect helicopter operating data. The built-in program of the processing host is used to implement a helicopter monitoring method based on mixed characteristic sound as described in any one of claims 1-5, and outputs monitoring information including early warning information, abnormal sound source location information, fault type and probability of the fault type.

7. A helicopter monitoring system based on hybrid characteristic sound according to claim 6, characterized in that: The processing host includes: an acoustic signal preprocessing module, an abnormal signal rapid detection module, an abnormal sound source localization module, a fault diagnosis module, and a data maintenance and update module; The acoustic signal preprocessing module is used to preprocess the acoustic data; The abnormal signal rapid detection module is used to extract at least one low-order feature data from the processed acoustic data, compare the low-order feature data with normal data in the database, and generate early warning information if an abnormality is found. The abnormal sound source localization module is used to obtain abnormal sound source localization information by means of the acoustic data, operational data, structural data and the prior probability of the fault distribution in the cabin when the acoustic data is abnormal. The fault diagnosis module is used to filter and enhance the acoustic data through the abnormal sound source localization information to obtain acoustic feature data to be identified, and input the acoustic feature data to be identified and the running data into the attention mechanism-based diagnostic model to obtain the fault type and the probability of the fault type. The data maintenance and update module is used to update the normal data of the database through the acoustic data and operational data, and to optimize the attention-based diagnostic model through the acoustic feature data to be identified, operational data, and actual fault types.

8. A helicopter monitoring system based on hybrid characteristic sound according to claim 7, characterized in that: The abnormal sound source localization module includes: The sound source spatial distribution probability calculation module is used to obtain the sound source spatial distribution probability through the acoustic data and operational data; The historical fault location distribution probability calculation module is used to obtain the historical fault location distribution probability through the structural data and the prior probability of fault distribution in the cabin. An abnormal sound source localization module is used to obtain abnormal sound source localization information through the spatial distribution probability of the sound source and the historical fault location distribution probability.

9. A helicopter monitoring system based on hybrid characteristic sound according to claim 7, characterized in that: The fault diagnosis module includes: The data filtering and enhancement module is used to filter acoustic data from the vicinity of the abnormal sound source location based on the abnormal sound source location information, and enhance the acoustic data in the direction of the abnormal sound source location to obtain acoustic feature data to be identified.

10. A helicopter monitoring system based on hybrid characteristic sound according to claim 9, characterized in that: The fault diagnosis module also includes: The heterogeneous feature fusion module is used to convert the acoustic feature data to be identified and the running data into vectors of the same dimension; The attention mechanism filtering module is used to extract feature values ​​from the vectors of the same dimension using an attention model and assign weights to the feature values. The fault identification module is used to input the weighted feature values ​​into the diagnostic model to obtain the classification result: the fault type and the probability of that fault type.

Images