CN-121980339-A - Deep learning driven outdoor hangar equipment fault prediction method and system

CN121980339ACN 121980339 ACN121980339 ACN 121980339ACN-121980339-A

Abstract

The invention discloses a deep learning driven outdoor hangar equipment fault prediction method and system, which comprises the steps of A1, collecting vibration frequency time sequence data, temperature-humidity time sequence data and environmental temperature-humidity time sequence data of a hangar part of an unmanned aerial vehicle, preprocessing, A2, respectively extracting vibration frequency preliminary time sequence characteristics and part temperature-humidity preliminary time sequence characteristics through a local characteristic enhancement convolution unit, A3, carrying out noise separation through a noise-fault distinguishing gating mechanism, A4, extracting early weak fault characteristics, A5, extracting environmental temperature-humidity time sequence characteristics, calculating fault development track characteristics, A6, calculating a hangar risk quantization value and judging risk grades, A7, judging a processing strategy corresponding to the current risk grade, generating fault early warning information adapting to the risk grade, and pushing grading reminding. The invention can solve the problem that the traditional method is difficult to realize fault pre-judgment in an outdoor complex scene.

Inventors

CHEN LECHUN
GUO QIANG
LI BINJIE

Assignees

江苏数字鹰科技股份有限公司

Dates

Publication Date: 20260505
Application Date: 20260108

Claims (8)

1. The deep learning driven outdoor hangar equipment fault prediction method is characterized by comprising the following steps of: A1, acquiring vibration frequency time sequence data, temperature-humidity time sequence data and environmental temperature-humidity time sequence data of an Unmanned Aerial Vehicle (UAV) component, and preprocessing to obtain preprocessed component vibration frequency time sequence data, preprocessed component temperature-humidity time sequence data and preprocessed environmental temperature-humidity time sequence data; A2, respectively extracting a vibration frequency preliminary time sequence characteristic and a component temperature-humidity preliminary time sequence characteristic by a local characteristic enhancement convolution unit according to the preprocessed component vibration frequency time sequence data and the preprocessed component temperature-humidity time sequence data; A3, performing noise separation through a noise-fault distinguishing and gating mechanism according to the initial time sequence characteristic of the vibration frequency, the initial time sequence characteristic of the temperature and the humidity of the component and the time sequence data of the preprocessed ambient temperature and the preprocessed ambient temperature, so as to obtain the time sequence characteristic of the vibration frequency after denoising and the time sequence characteristic of the temperature and the humidity of the component after denoising; A4, merging the denoised vibration frequency time sequence characteristics and the denoised component temperature-humidity time sequence characteristics to obtain early weak fault characteristics; a5, extracting the environmental temperature-humidity time sequence characteristics according to the preprocessed environmental temperature-humidity time sequence data and the early weak fault characteristics, and calculating fault development track characteristics through an improved LSTM network; A6, calculating early weak fault feature weights and fault development track feature weights according to the early weak fault features and the fault development track features, and fusing the early weak fault feature weights and the fault development track feature weights to obtain fault fusion features; a7, judging a processing strategy corresponding to the current risk level according to the risk quantification value of the unmanned aerial vehicle hangar and the risk level of the unmanned aerial vehicle hangar, generating fault early warning information adapting to the risk level according to the processing strategy corresponding to the current risk level, and pushing grading reminding.
2. The deep learning driven outdoor hangar equipment failure prediction method according to claim 1, wherein the A1 step includes: A11, acquiring vibration frequency time sequence data of an unmanned aerial vehicle hangar component through a piezoelectric vibration sensor, wherein the data type is time sequence waveform data, and the data comprises motor rotating shaft vibration frequency and telescopic mechanism guide rail vibration frequency; A12, acquiring temperature-humidity time sequence data of the unmanned aerial vehicle library component through a platinum resistance temperature sensor, wherein the data type is time sequence numerical data comprising motor winding temperature and telescopic mechanism driving module temperature, acquiring temperature-humidity time sequence data of the environment through a temperature and humidity sensor, and the data type is time sequence numerical data; A13, performing time stamp alignment and uniform data length on the time sequence data of the vibration frequency of the part after preliminary pretreatment, the time sequence data of the temperature of the part after preliminary pretreatment and the time sequence data of the ambient temperature and humidity after preliminary pretreatment to obtain the time sequence data of the vibration frequency of the part after pretreatment, the time sequence data of the temperature and humidity of the part after pretreatment and the time sequence data of the ambient temperature and humidity after pretreatment.
3. The deep learning driven outdoor hangar equipment failure prediction method according to claim 1, wherein the calculation method of the A2 step includes: Wherein, the For the preliminary timing characteristics of the vibration frequency, The convolution unit is enhanced for the local feature, For the Hadamard product, As a function of the ReLU, For the purpose of batch normalization, For the pre-processed component vibration frequency timing data, For the component temperature-humidity preliminary timing characteristics, For the pre-processed component temperature-humidity time series data, For the summation from element to element, Pooling for global averaging; The calculation mode of the local characteristic enhancement convolution unit is as follows: Wherein, the Is a one-dimensional convolution layer which is formed by a plurality of layers, The input of the convolution unit is enhanced for the local feature, For an adaptive adjustment factor of the local receptive field at x, As a function of the Sigmoid, For the purpose of the average pooling, The weight matrix is adaptively adjusted for the local receptive field, The bias vector is adaptively adjusted for the local receptive field.
4. The deep learning driven outdoor hangar equipment failure prediction method according to claim 3, wherein the calculation mode of the noise-failure distinguishing gating mechanism in the step A3 includes: Wherein, the Gating weights are distinguished for noise-faults, Is a multi-layer sensing machine, which is a multi-layer sensing machine, In order for the splicing operation to be performed, As the average value of the environmental temperature-humidity time series data after pretreatment, For the de-noised vibration frequency timing characteristics, Is the temperature-humidity time sequence characteristic of the component after denoising.
5. The deep learning driven outdoor hangar equipment failure prediction method according to claim 4, wherein the calculation method of the A4 step includes: Wherein, the In order to be a feature of early weak failure, For the purpose of the L2 regularization, For the vibration frequency weight matrix, For the component temperature-humidity weight matrix, Is the mechanism of attention.
6. The deep learning driven outdoor hangar equipment failure prediction method according to claim 5, wherein the A5 step includes: A51, calculating the difference value of adjacent time steps according to the preprocessed environmental temperature-humidity time sequence data, combining according to a time sequence to obtain the environmental temperature-humidity time sequence difference value data, and extracting the environmental temperature-humidity time sequence characteristics by combining with early weak fault characteristics, wherein the calculation mode is as follows: Wherein, the For the preliminary ambient temperature-humidity timing characteristics, For the pre-processed ambient temperature-humidity time series data, For the purpose of maximum pooling, In order to gate the circulation unit, Is the time sequence difference data of the ambient temperature and the humidity, For the context-fault associated weight matrix, Is an ambient temperature-humidity timing characteristic; A52, inputting the preprocessed environment temperature-humidity time sequence data into an improved LSTM network for time sequence evolution modeling, and calculating fault development track characteristics in the following calculation modes: Wherein, the In order for the fault to develop a wave characteristic, For the adaptive correction coefficient of the environmental fluctuations, For the covariance calculation the sum of the values of the covariance, In order to take the absolute value of the value, To avoid a minimum value with a denominator of 0, In order to be in a fault hidden state, Is a long-term and short-term memory network, Is a fault development track feature.
7. The deep learning driven outdoor hangar equipment failure prediction method according to claim 6, wherein the A6 step includes: A61, calculating early weak fault feature weights and fault development track feature weights according to the early weak fault features and the fault development track features; a62, fusing the early weak fault characteristics and the fault development track characteristics based on the early weak fault characteristic weight and the fault development track characteristic weight to obtain fault fusion characteristics; and A63, calculating a risk quantification value of the unmanned aerial vehicle hangar based on the fault fusion characteristics and judging the risk grade of the unmanned aerial vehicle hangar.
8. An outdoor hangar equipment fault prediction system driven by deep learning, which is characterized by comprising: the data acquisition module acquires vibration frequency time sequence data, temperature-humidity time sequence data and environmental temperature-humidity time sequence data of the unmanned aerial vehicle library component and performs preprocessing to obtain preprocessed component vibration frequency time sequence data, preprocessed component temperature-humidity time sequence data and preprocessed environmental temperature-humidity time sequence data; the feature extraction module is used for respectively extracting the initial time sequence feature of the vibration frequency and the initial time sequence feature of the temperature-humidity of the component through the local feature enhancement convolution unit according to the preprocessed time sequence data of the vibration frequency of the component and the preprocessed time sequence data of the temperature-humidity of the component; the characteristic denoising module performs noise separation through a noise-fault distinguishing and gating mechanism according to the initial time sequence characteristic of the vibration frequency, the initial time sequence characteristic of the temperature and the humidity of the component and the time sequence data of the preprocessed ambient temperature and the preprocessed ambient humidity to obtain the denoised vibration frequency time sequence characteristic and the denoised temperature and humidity time sequence characteristic of the component; The feature fusion module is used for fusing the denoised vibration frequency time sequence feature and the denoised component temperature-humidity time sequence feature to obtain an early weak fault feature; The fault development track feature extraction module is used for extracting the environmental temperature-humidity time sequence features according to the preprocessed environmental temperature-humidity time sequence data and the early weak fault features; the risk quantification module is used for calculating early weak fault feature weights and fault development track feature weights according to the early weak fault features and the fault development track features, and fusing the early weak fault feature weights and the fault development track feature weights to obtain fault fusion features; the early warning module judges a processing strategy corresponding to the current risk level according to the risk quantification value of the unmanned aerial vehicle hangar and the risk level of the unmanned aerial vehicle hangar, generates fault early warning information adapting to the risk level according to the processing strategy corresponding to the current risk level, and pushes grading reminding so as to realize the deep learning-driven outdoor hangar equipment fault prediction method according to any one of claims 1-7.

Description

Deep learning driven outdoor hangar equipment fault prediction method and system Technical Field The invention relates to the technical field of unmanned aerial vehicle hangars, in particular to a deep learning driven outdoor hangar equipment fault prediction method and system. Background The unmanned aerial vehicle outdoor hangar equipment is used as a core infrastructure of an unmanned aerial vehicle operation and maintenance system, the stable operation of the unmanned aerial vehicle outdoor hangar equipment directly influences operation and maintenance efficiency and operation continuity, the outdoor hangar equipment fault prediction technology is a technology system for predicting potential faults of key parts of equipment by collecting various parameters in the operation process of the equipment and combining a specific analysis method, and aims to identify fault precursors in advance and trigger maintenance intervention and avoid operation and maintenance interruption caused by sudden shutdown. The traditional unmanned aerial vehicle hangar equipment fault prevention and control method mainly relies on two modes of regular preventive maintenance and post-maintenance, wherein the regular preventive maintenance sets a fixed maintenance period according to equipment operation time length or experience, the method cannot adapt to the differential influence of complex working conditions such as temperature fluctuation, humidity change and the like on equipment parts in an outdoor environment, the problem of excessive maintenance or untimely maintenance is easy to occur, and the post-maintenance is to overhaul after equipment is in fault shutdown, so that unmanned aerial vehicle operation and maintenance operation is interrupted, and operation and maintenance cost and shutdown loss are increased. With the application expansion of deep learning technology in the field of fault prediction, a part of schemes begin to analyze equipment operation data by adopting a deep learning model to realize fault prediction, the prior art is mostly based on Convolutional Neural Network (CNN), cyclic neural network (RNN) and other models to perform feature extraction and fault recognition on single type equipment operation parameters, or develop fault prediction research aiming at equipment in indoor fixed environment, however, the problems of strong time sequence relevance of core component operation parameters, large parameter fluctuation caused by outdoor complex environment and the like in the long-term operation process of outdoor hangar equipment exist, the defects of insufficient time sequence feature mining, insufficient recognition sensitivity of weak fault precursors in complex environment and difficult fault risk level pre-judgment exist in the existing deep learning scheme, and the core pain points of the critical component fault precursors of the outdoor hangar equipment, which influence the operation and maintenance efficiency due to sudden shutdown cannot be effectively solved, so that a deep learning driving fault prediction technology capable of accurately capturing multi-dimensional time sequence operation parameter relevance features and adapting to outdoor complex working conditions is needed to improve the accuracy and timeliness of fault pre-judgment. Disclosure of Invention In view of the above, the present invention aims to provide a deep learning driven outdoor hangar equipment fault prediction method and system, so as to solve the problem that the conventional method is difficult to implement fault prediction in an outdoor complex scene. A deep learning driven outdoor hangar equipment fault prediction method comprises the following steps: A1, acquiring vibration frequency time sequence data, temperature-humidity time sequence data and environmental temperature-humidity time sequence data of an Unmanned Aerial Vehicle (UAV) component, and preprocessing to obtain preprocessed component vibration frequency time sequence data, preprocessed component temperature-humidity time sequence data and preprocessed environmental temperature-humidity time sequence data; A2, respectively extracting a vibration frequency preliminary time sequence characteristic and a component temperature-humidity preliminary time sequence characteristic by a local characteristic enhancement convolution unit according to the preprocessed component vibration frequency time sequence data and the preprocessed temperature-humidity time sequence data; A3, performing noise separation through a noise-fault distinguishing and gating mechanism according to the initial time sequence characteristic of the vibration frequency, the initial time sequence characteristic of the temperature and the humidity of the component and the time sequence data of the preprocessed ambient temperature and the preprocessed ambient temperature, so as to obtain the time sequence characteristic of the vibration frequency after denoising and the time sequence chara