WO2022228049A1 - 基于5g边缘计算和深度学习的航空发动机故障诊断方法 - Google Patents

基于5g边缘计算和深度学习的航空发动机故障诊断方法 Download PDF

Info

Publication number
WO2022228049A1
WO2022228049A1 PCT/CN2022/085126 CN2022085126W WO2022228049A1 WO 2022228049 A1 WO2022228049 A1 WO 2022228049A1 CN 2022085126 W CN2022085126 W CN 2022085126W WO 2022228049 A1 WO2022228049 A1 WO 2022228049A1
Authority
WO
WIPO (PCT)
Prior art keywords
aero
engine
data
layer
neural network
Prior art date
Application number
PCT/CN2022/085126
Other languages
English (en)
French (fr)
Inventor
万安平
杨洁
袁建涛
王景霖
单添敏
Original Assignee
浙大城市学院
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 浙大城市学院 filed Critical 浙大城市学院
Publication of WO2022228049A1 publication Critical patent/WO2022228049A1/zh

Links

Images

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00Computer-aided design [CAD]
    • G06F30/20Design optimisation, verification or simulation
    • G06F30/27Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00Computing arrangements based on biological models
    • G06N3/02Neural networks
    • G06N3/04Architecture, e.g. interconnection topology
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00Computing arrangements based on biological models
    • G06N3/02Neural networks
    • G06N3/08Learning methods
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F2111/00Details relating to CAD techniques
    • G06F2111/10Numerical modelling
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F2119/00Details relating to the type or aim of the analysis or the optimisation
    • G06F2119/04Ageing analysis or optimisation against ageing

Definitions

  • the invention relates to the field of complex equipment fault diagnosis, in particular to the field of aero-engine fault diagnosis, and in particular to an aero-engine fault diagnosis method based on 5G edge computing and deep learning.
  • the fault diagnosis and identification of aero-engine is mainly to classify and predict the fault categories of rotating machinery such as gears and bearings.
  • the vibration signal analysis method is the most widely used research method in the fault diagnosis of aero-engine gears and bearings. By collecting the vibration acceleration signals during the working process of gears and bearings with different damage conditions, machine learning methods are used to classify and predict the signals and find faults. The potential characteristics of data greatly improve the efficiency and accuracy of fault diagnosis.
  • the purpose of the present invention is to provide an aero-engine fault diagnosis method based on 5G edge computing and deep learning to overcome the deficiencies of the prior art.
  • the present invention provides the following scheme:
  • An aero-engine fault diagnosis method based on 5G edge computing and deep learning comprising the following steps:
  • Step 1 Data collection, preprocessing and storage based on the new 5G cloud-edge-terminal network architecture
  • Step 1.2 establish aero-engine fault database management system; preprocess and store data;
  • Step 2 building a machine learning module in the edge cloud, the historical data stored in the aero-engine fault database management system provides training samples for the machine learning module;
  • the machine learning module utilizes the data from the aero-engine fault database management system, Predict and infer aero-engine behavior through a one-dimensional convolutional neural network model, and jointly optimize the allocation of communication and computing resources;
  • Step 2.1 build a one-dimensional convolutional neural network model;
  • the one-dimensional convolutional neural network model includes an input layer, five convolutional layers, five pooling layers, a fully connected layer and an output layer;
  • the convolutional neural network model performs feature extraction and type identification on the vibration signal, and finally outputs the probability value of the vibration signal of each fault category as the identification result;
  • Step 2.2 One-dimensional convolutional neural network model training and result visualization: Input the processed aero-engine vibration signal into the one-dimensional convolutional neural network to be trained, set the ratio of training set and test set, the number of model iterations, and the single Train the input data batch, training batch and network parameters, and monitor the recognition accuracy and loss function value of the one-dimensional convolutional neural network model in real time; output the recognition effect of the one-dimensional convolutional neural network model in a visual way;
  • Step 2.3 The realization model of joint optimal allocation of resources adopted in joint optimal allocation of communication and computing resources is as follows:
  • is the maximum error tolerance probability of data packets of ultra-reliable and low-latency communication services
  • D is the actual delay experienced by the data packet transmission
  • P Suc (D>D threshold ) ⁇ is the probability delay constraint
  • is the excess time delay. Proportion of resources for high-reliability and low-latency communication services
  • Step 3 Intelligent self-management of aero-engine gear fault simulation platform and aero-engine fault database management system: a decision-making center is designed inside the aero-engine gear fault simulation platform, and the decision-making center accepts the output from the machine learning The results of the modular machine learning are analyzed and made decisions; the decision center manages the aero-engine fault database management system at the same time, and instructs the aero-engine fault database management system to cache in advance.
  • the base station in step 1.1 ensures that the terminal equipment in each aero-engine group satisfies the delay constraint of service transmission by configuring the number of time slots K included:
  • T I is the duration of the terminal device's initial data transmission
  • T R is the duration of the re-transmission of data after the terminal device's initial data transmission error
  • T n (1 ⁇ n ⁇ N Tot , n ⁇ N + ) is the terminal device n's initial transmission time The time interval between the moment when the data fails and the next time the data is resent
  • N Tot is the total number of terminal devices in a terminal device group
  • T Threshold is the delay constraint of service transmission;
  • the base station allocates independent initial data transmission resources to each terminal device according to the number of terminal devices in the same group; after the initial data transmission fails, the terminal devices in the same group wait for the base station to configure the retransmission resource time, and then retransmit the data;
  • the number of time slots included K is the size of each packet; the number of time slots K is set as:
  • T Threshold is the delay constraint of service transmission
  • T S is the length of a data transmission time slot
  • TI is the duration of the terminal equipment to transmit data for the first time
  • T R is the duration of the terminal equipment to retransmit data after the initial data transmission error.
  • step 1.2 specifically includes the following steps:
  • Step 1.2.1 the aero-engine fault database management system interacts with the aero-engine and the cloud, accepts the data from the aero-engine and caches the data, and uploads the data to the cloud;
  • Step 1.2.2 Perform missing value processing, abnormal value processing and normalization processing on the voltage signal corresponding to the original vibration signal collected in step 1.1, and mark the data of different fault types.
  • the missing value processing in step 1.2.2 is to complete the missing value by the average value of the values on both sides of the missing value; the abnormal value processing is to discard the abnormal value; the normalization processing adopts the maximum and minimum value normalization, normalization
  • the formula is:
  • xmax is the maximum value of the sample data
  • xmin is the minimum value of the sample data
  • x ⁇ is the normalization result
  • the numerical interval of the sample data is [0,1].
  • the input layer of the one-dimensional convolutional neural network model in step 2.1 is connected to five convolutional layers respectively, the five convolutional layers are respectively connected to five pooling layers, and the five pooling layers are aggregated and connected to the fully connected layer.
  • the connection layer connects the output layer.
  • the input layer feature map group of the one-dimensional convolutional neural network model in step 2.1 is a two-dimensional tensor, wherein each slice one-dimensional array is an input feature map, and the number of channels of the input layer is aero-engine gear failure simulation.
  • three neurons are connected between every two layers of neural networks in the convolutional layer of the one-dimensional convolutional neural network model in step 2.1, and the formula for extracting local area features during the convolution process is:
  • the pooling layer of the one-dimensional convolutional neural network model in step 2.1 adopts the maximum pooling method, and selects the maximum activity value of all neurons in the pooling area as the representation of the pooling area, and the expression of the pooling function.
  • the formula is:
  • xi is the activity value of each neuron in the pooling region Rd .
  • the fully connected layer of the one-dimensional convolutional neural network model in step 2.1 non-linearly combines the features extracted by the convolutional layer and the pooling layer:
  • the visualization method in step 2.2 is a line graph.
  • the present invention discloses the following technical effects:
  • the invention efficiently utilizes the limited aero-engine fault data resources under the 5G emerging network architecture, combined with computing and storage resources, can improve the storage and transmission speed of aero-engine massive operation data, and provide a reliable basis for aero-engine fault identification.
  • the invention uses machine learning technology to build an intelligent learning module in the edge cloud.
  • the original vibration data of aero-engine gears one-dimensional time domain vibration signal
  • Input through layer-by-layer feature extraction, to complete the identification of fault types
  • edge computing technology processes data close to the network access end, reducing data transmission costs, saving time and improving efficiency
  • one-dimensional convolutional neural network (1D-CNN) It can directly perform feature mining on the time-domain signal, and the collected one-dimensional time-domain vibration signal is used as the sample space input network, which eliminates the original signal processing process, and can complete the fault type identification and diagnosis more quickly. processing has potential application value.
  • the present invention directly performs the convolutional neural network operation on the original one-dimensional vibration signal, the process is relatively simple, the rich signal processing expert experience is not required, and the recognition effect is also ideal.
  • Fig. 1 is the flow chart of the aero-engine fault diagnosis method based on 5G edge computing and deep learning of the present invention
  • Figure 2 is a data diagram of aero-engine gear fault signal
  • Figure 3 is a schematic diagram of a one-dimensional convolutional neural network structure
  • Figure 4 is a schematic diagram of the visualization of the training results of the one-dimensional convolutional neural network model.
  • This embodiment provides an aero-engine fault diagnosis method based on 5G edge computing and deep learning, the process of which is shown in Figure 1, and specifically includes the following steps:
  • Step 1 Data collection, preprocessing and storage based on the new 5G cloud-edge-terminal network architecture.
  • Data collection Build an aero-engine gear fault simulation platform, adopt edge computing technology (5G core technology), and arrange base stations in the edge network close to the aero-engine gear fault simulation platform for data collection, and data is directly processed at the edge of the network. Processing, transmission and storage to avoid the delay and loss caused by the data returning to the core network 2; the acceleration sensor installed on the aero-engine gear fault simulation platform collects the vibration signals of different types of gears in different positions and directions, and converts the vibration signals into Voltage signal; the base station ensures that the terminal equipment in each aero-engine unit meets the delay constraint of service transmission by configuring the number of time slots K included:
  • T I is the duration of the terminal device's initial data transmission
  • T R is the duration of the re-transmission of data after the terminal device's initial data transmission error
  • T n (1 ⁇ n ⁇ N Tot , n ⁇ N + ) is the terminal device n's initial transmission time The time interval between the moment when the data fails and the moment when the data is resent next time
  • N Tot is the total number of terminal devices in a terminal device group
  • T Threshold is the delay constraint of service transmission.
  • the base station allocates independent initial data transmission resources for each terminal device according to the number of terminal devices in the same group; after the initial data transmission fails, the terminal devices in the same group wait for the base station to configure the retransmission resource time, and then retransmit data.
  • the number of time slots included K is the size of each packet; the number of time slots K is set as:
  • T Threshold is the delay constraint of service transmission
  • T S is the length of a data transmission time slot
  • TI is the duration of the terminal equipment to transmit data for the first time
  • T R is the duration of the terminal equipment to retransmit data after the initial data transmission error.
  • Step 1.2 The data acquisition system uses database technologies, such as mySQL, Oracle, etc., to establish an aero-engine fault database management system to realize data interaction and effective storage; preprocess and store data.
  • database technologies such as mySQL, Oracle, etc.
  • Step 1.2.1 The aero-engine fault database management system interacts with the aero-engine and the cloud, accepts the data from the aero-engine, caches the data for the aero-engine in advance, and uploads the data to the cloud.
  • Step 1.2.2 Perform missing value processing, outlier processing and normalization processing on the voltage signal corresponding to the original vibration signal collected in step 1.1, and mark the data of different fault types to ensure the efficiency of subsequent model training.
  • the missing value processing is to complete the missing value by the average value of the two sides of the missing value;
  • the abnormal value processing is to discard the abnormal value (the abnormal value is the sudden and transient fluctuation of the data caused by the instrument error);
  • the normalization processing adopts the maximum and minimum values.
  • Unification, the normalization formula is:
  • xmax is the maximum value of the sample data
  • xmin is the minimum value of the sample data
  • x ⁇ is the normalization result
  • the numerical interval of the sample data is [0,1].
  • Step 2 Use machine learning technology to build a machine learning module in the edge cloud, and the historical data stored in the aero-engine fault database management system provides training samples for the machine learning module; the machine learning module uses the data from the aero-engine fault database management system.
  • the dimensional convolutional neural network model predicts and infers aero-engine behavior, and jointly optimizes the allocation of communication and computing resources.
  • Step 2.1 Use Python language to build a one-dimensional convolutional neural network model;
  • the one-dimensional convolutional neural network model includes an input layer, five convolutional layers, five pooling layers, a fully connected layer and an output layer (specific network). The structure can be adjusted according to specific data);
  • the input layer of the one-dimensional convolutional neural network model is connected to five convolutional layers respectively, the five convolutional layers are respectively connected to five pooling layers, and the five pooling layers are aggregated and connected to the full connection
  • the fully connected layer is connected to the output layer;
  • the feature extraction and type identification of the vibration signal is carried out through the one-dimensional convolutional neural network model, and the probability value of the vibration signal of each fault category is finally output as the identification result;
  • the input layer feature map group is a two-dimensional tensor, in which each slice one-dimensional array is an input feature map, and the number of channels in the input layer is the number of acceleration sensors installed on the aero-engine gear fault simulation platform;
  • the vibration signals at different positions and directions of the gears are respectively set as data features, and the fault category is set as data labels.
  • x l (j) represents the input of the j-th neuron in the l-th layer, represents the input of the jth neuron in layer l+1, At the same time, it is the output of the l layer;
  • f() is a nonlinear activation function, generally the Relu function is used, and the symbol * represents the dot product of the kernel and the local area; after each convolution operation is completed, the output feature is mapped to a one-dimensional tensor .
  • the pooling layer adopts the maximum pooling method, and selects the maximum activity value of all neurons in the pooling area as the representation of the pooling area.
  • the expression of the pooling function is:
  • xi is the activity value of each neuron in the pooling area Rd ; the division of each feature map in the pooling layer should not be too large, and the pooling kernel is set to 2 ⁇ 1.
  • the fully connected layer performs a nonlinear combination of the features extracted by the convolutional layer and the pooling layer:
  • Step 2.2 Training of the one-dimensional convolutional neural network model and visualization of the results: Input the above-processed aero-engine vibration signal into the one-dimensional convolutional neural network to be trained, and set the ratio of the training set to the test set (usually 4:1). , when the amount of data is large, the proportion of the test set can be appropriately increased), the number of model iterations (500 times, and can also be adjusted appropriately according to the amount of data), the batch of data sent to a single training, the training batch and the network parameters (32 samples, or an integer multiple of 16), and monitor the recognition accuracy of the one-dimensional convolutional neural network model and the change of the loss function value in real time; output the recognition effect of the one-dimensional convolutional neural network model in the form of a line graph.
  • Step 2.3 The realization model of joint optimal allocation of resources adopted in joint optimal allocation of communication and computing resources is as follows:
  • is the maximum error tolerance probability of data packets of the ultra-reliable and low-latency communication (URLLC) service
  • D is the actual delay experienced by the data packet transmission
  • P Suc (D>D threshold ) ⁇ is the probability delay constraint
  • n is the resource ratio of the ultra-reliable and low-latency communication (URLLC) service; when the URLLC service and the enhanced mobile broadband (eMBB) service are multiplexed in the resource reservation mode, the larger the URLLC resource ratio n is, the URLLC service The better the quality of service, but the greater the impact on the eMBB business, so it is necessary to minimize ⁇ under the constraint of ensuring the high quality of service (QoS) of URLLC, so that it is the smallest correspondingly to the eMBB business.
  • QoS quality of service
  • Step 3 Intelligent self-management of aero-engine gear fault simulation platform and aero-engine fault database management system: a decision-making center is designed inside the aero-engine gear fault simulation platform, and the decision-making center accepts the output from the machine learning module, and uses game theory and other tools to analyze the machine The results of machine learning in the learning module are analyzed and decided to realize functions such as computing offloading, edge-cloud collaboration, and resource optimal allocation; the decision-making center also conducts intelligent management of the aero-engine fault database management system and instructs the aero-engine fault database management system to cache in advance.
  • Example 1 On the basis of Example 1, as shown in Figure 2, five different fault types (normal gear (a), broken tooth (b), missing tooth (c) under the same sensor arrangement collected for the aero-engine gear fault simulation platform , tooth surface wear (d), tooth root crack (e), the vibration signal data can be added and deleted according to the specific situation) (the horizontal axis is the sampling time, the vertical axis is the amplitude signal collected by the acceleration sensor converted into a voltage value),
  • the data can also be real-time status data collected by the aero-engine under real operating conditions.
  • the sampling frequency and sensor location arrangement set during data acquisition can be determined according to the actual situation. If necessary, in order to improve the accuracy of model training, other types of sensors such as acoustic sensors can be added.
  • Figure 3 is a schematic diagram of the structure of the constructed one-dimensional convolutional neural network.
  • the specific structure of the one-dimensional convolutional neural network model proposed in this embodiment consists of five convolutional layers, five pooling layers, one fully connected layer and one Softmax output layer composition. After going through the first convolutional layer, the signal is transformed into a set of feature maps, which are then downsampled by max pooling. After these operations are repeated 4 times, the features of the last pooling layer are connected to the fully connected layer, then the fully connected layer is activated through the Relu function, and passed to the Softmax layer, and finally the probability value of each classification is obtained, among which the class with the highest probability will be regarded as the recognition result.
  • a dropout layer needs to be added, and the dropout rate is 20%, that is, 20% of the training parameters are dropped, which improves the training speed of the model and prevents overfitting.
  • the specific parameters of the network are set as shown in Table 2 below.
  • the preprocessed data needs to be input into the neural network according to the ratio of training set/test set to 4/1, the number of model iterations is set to 100, the training batch is 32 each time, and the final drawing is accurate through matplotlib
  • the real-time changes of the rate and loss function values (as shown in Figure 4) can further draw the confusion matrix of the recognition results and observe the specific recognition situation of the model.
  • the parameters of the one-dimensional convolutional neural network model can be gradually adjusted to optimize the training effect.
  • the present embodiment directly performs the convolutional neural network operation on the original one-dimensional vibration signal, the process is relatively simple, does not require rich signal processing expert experience, the recognition effect is also ideal, and the minimum recognition rate can reach 78.97%.
  • Table 3 below shows a comparison result table between the method of this embodiment and the traditional machine learning method.
  • the recognition accuracy of 63.90% is increased by 15.07%, and the accuracy of the support vector machine is increased by 15.89%, and the root mean square error of gear fault diagnosis is better. Low. Further parameter optimization for specific data can more accurately identify aero-engine faults.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Evolutionary Computation (AREA)
  • Software Systems (AREA)
  • Artificial Intelligence (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Health & Medical Sciences (AREA)
  • Mathematical Physics (AREA)
  • Computational Linguistics (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • Data Mining & Analysis (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Medical Informatics (AREA)
  • Computer Hardware Design (AREA)
  • Geometry (AREA)
  • Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)

Abstract

本发明涉及一种基于5G边缘计算和深度学习的航空发动机故障诊断方法,包括步骤:基于5G新型云边端网络架构的数据采集、预处理及存储;在边缘云中构建机器学习模块,航空发动机故障数据库管理系统存储的历史数据为机器学习模块提供训练样本;航空发动机齿轮故障模拟平台和航空发动机故障数据库管理系统智能化自我管理。本发明的有益效果是:利用5G新兴网络架构下有限的航空发动机故障数据资源,结合计算和存储资源,能够提高航空发动机海量运行数据的存储与传输速度,为航空发动机故障识别提供可靠的依据;本发明直接对原始一维振动信号进行卷积神经网络操作,过程较为简易,不需要丰富的信号处理专家经验,识别效果也较为理想。

Description

基于5G边缘计算和深度学习的航空发动机故障诊断方法
本申请要求于2021年04月27日提交中国专利局、申请号为202110457105.X、发明名称为“一种基于5G边缘计算和深度学习的航空发动机故障诊断方法”的中国专利申请的优先权,其全部内容通过引用结合在本申请中。
技术领域
本发明涉及复杂设备故障诊断领域,尤其涉及航空发动机故障诊断领域,特别涉及一种基于5G边缘计算和深度学习的航空发动机故障诊断方法。
背景技术
自20世纪末至今,伴随5G信息技术的不断发展,人工神经网络由于其具有强大的并行处理能力、非线性函数逼近能力和自组织、自学习、自适应等特点,在航空航天领域的应用愈发广泛,已然成为现阶段航空飞行器故障诊断的关键手段之一。
航空发动机作为航空飞行器最重要的动力组成部分,因其机械结构复杂、工作环境恶劣,在使用较长时间后其内部零件容易产生机械损伤,使得工作性能会大幅度降低,例如,构成其旋转机构的轴系零件(如齿轮、轴承等),在表面磨损等失效问题下,容易导致发动机组件产生巨大的振动和噪声、降低运行效率,严重还会引起整个机组的破坏,造成巨大的经济损失。如果未能实时准确地检测故障的发生,将对空中作业的安全性和效率产生巨大隐患。因此,对航空发动机进行状态监测并及时准确地判断故障类别并预测故障发生对于保证飞行安全具有重要意义。
航空发动机的故障诊断识别主要是对旋转机械如齿轮和轴承的故障类别进行特挖掘分类和预测。振动信号分析法是航空发动机齿轮和轴承故障诊断中应用最为广泛的研究方法,通过采集不同损伤情况的齿轮和轴承工作过程中的振动加速度信号,应用机器学习手段对信号进行分类与预测,挖掘故障数据的潜在特征,对故障诊断效率和准确率有着极大的提升。
发明内容
本发明的目的是提供一种基于5G边缘计算和深度学习的航空发动机故障 诊断方法,以克服现有技术的不足。
为实现上述目的,本发明提供了如下方案:
一种基于5G边缘计算和深度学习的航空发动机故障诊断方法,包括以下步骤:
步骤1、基于5G新型云边端网络架构的数据采集、预处理及存储;
步骤1.1、数据采集:搭建航空发动机齿轮故障模拟平台,采用边缘计算技术,在靠近所述航空发动机齿轮故障模拟平台的边缘网络布置基站;通过所述航空发动机齿轮故障模拟平台安装的加速度传感器采集齿轮不同位置和方向的振动信号,将振动信号转换为电压信号;
步骤1.2、建立航空发动机故障数据库管理系统;对数据进行预处理和存储;
步骤2、在边缘云中构建机器学习模块,所述航空发动机故障数据库管理系统存储的历史数据为机器学习模块提供训练样本;所述机器学习模块利用来自所述航空发动机故障数据库管理系统的数据,通过一维卷积神经网络模型对航空发动机行为进行预测和推理,对通信和计算资源进行联合优化分配;
步骤2.1、搭建一维卷积神经网络模型;一维卷积神经网络模型包括一个输入层,五个卷积层,五个池化层,一个全连接层和一个输出层;通过所述一维卷积神经网络模型对振动信号进行特征提取和类型识别,最终输出各故障类别振动信号的概率值作为识别结果;
步骤2.2、一维卷积神经网络模型的训练及结果可视化:将处理后的航空发动机振动信号输入待训练的一维卷积神经网络,设置训练集和测试集的比例、模型迭代次数、单次训练送入数据批量、训练批次和网络参数,并实时监测一维卷积神经网络模型的识别准确率和损失函数值的变化;以可视化方式输出一维卷积神经网络模型的识别效果;
步骤2.3、对通信和计算资源进行联合优化分配时采用的资源联合优化分配实现模型如下:
min η
s.t.P Suc(D>D threshold)≤ε
其中,ε为超高可靠低延时通信业务的数据包最大容忍错误概率,D为数 据包传输经历的实际时延,P Suc(D>D threshold)≤ε为概率时延约束,η为超高可靠低延时通信业务的资源占比;
步骤3、航空发动机齿轮故障模拟平台和航空发动机故障数据库管理系统智能化自我管理:在所述航空发动机齿轮故障模拟平台内部设计决策中心,决策中心接受来自所述机器学习模块的输出,对机器学习模块机器学习的结果进行分析与决策;决策中心同时对所述航空发动机故障数据库管理系统进行管理,指导所述航空发动机故障数据库管理系统进行提前缓存。
可选地,步骤1.1中的基站通过配置包含的时隙数目K来保证每个航空发动机组内的终端设备都满足业务传输的时延约束:
Figure PCTCN2022085126-appb-000001
其中,T I是终端设备初次传输数据的时长;T R是终端设备初次数据传输出错后重新传输数据的时长;T n(1≤n≤N Tot,n∈N +)是终端设备n初次传输数据失败的时刻到下一次重新发送数据时刻之间的时间间隔;N Tot是一个终端设备分组中的总的终端设备数;T Threshold是业务传输的时延约束;
基站按照同一个分组内的终端设备数目为每个终端设备分配独立的初次数据传输资源;同一个分组内的终端设备在初次数据传输失败后,等待基站配置重传资源时刻,再重新传输数据;
包含的时隙数目K为每个分组的大小;时隙数目K设置为:
Figure PCTCN2022085126-appb-000002
其中,T Threshold是业务传输的时延约束,T S是一个数据传输时隙的长度,T I是终端设备初次传输数据的时长;T R是终端设备初次数据传输出错后重新传输数据的时长。
可选地,步骤1.2具体包括以下步骤:
步骤1.2.1、所述航空发动机故障数据库管理系统与航空发动机和云端进行数据交互,接受来自航空发动机的数据并缓存数据,向云端上传数据;
步骤1.2.2、将步骤1.1采集得到的原始振动信号对应的电压信号进行缺失值处理、异常值处理和归一化处理,并对不同故障类型的数据进行标记。
可选地,步骤1.2.2中缺失值处理为通过缺失值两侧值的平均值补全缺失值;异常值处理为舍弃异常值;归一化处理采用最大最小值归一化,归一化公 式为:
Figure PCTCN2022085126-appb-000003
其中,xmax为样本数据的最大值,xmin为样本数据的最小值,x`为归一化结果,样本数据的数值区间为[0,1]。
可选地,步骤2.1中一维卷积神经网络模型的输入层分别连接五个卷积层,五个卷积层分别连接五个池化层,五个池化层汇总连接全连接层,全连接层连接输出层。
可选地,步骤2.1中一维卷积神经网络模型的输入层特征映射组为二维张量,其中每个切片一维数组为一个输入特征映射,输入层的通道数为航空发动机齿轮故障模拟平台安装的加速度传感器数量;将传感器采集的不同故障类型的齿轮不同位置和方向的振动信号分别设置为数据特征,将故障类别设置为数据标签。
可选地,步骤2.1中一维卷积神经网络模型的卷积层中每两层神经网络之间连接三个神经元,卷积过程中局部区域特征提取的公式为:
Figure PCTCN2022085126-appb-000004
其中,
Figure PCTCN2022085126-appb-000005
表示第i个卷积核在l层的权重,
Figure PCTCN2022085126-appb-000006
表示第i个卷积核在l层的偏置,x l(j)表示第l层的第j个神经元的输入,
Figure PCTCN2022085126-appb-000007
表示第j个神经元在l+1层的输入,
Figure PCTCN2022085126-appb-000008
同时为l层的输出;f()为非线性激活函数,符号*表示内核与该局部区域的点积;每次卷积运算完成后,将输出特征映射为一维张量。
可选地,步骤2.1中一维卷积神经网络模型的池化层采取最大值池化方法,选择该池化区域所有神经元的最大活性值作为这个池化区域的表示,池化函数的表达式为:
y d=max x i(i∈R d)
其中,x i为池化区域R d内每个神经元的活性值。
可选地,步骤2.1中一维卷积神经网络模型的全连接层对卷积层和池化层提取的特征进行非线性组合:
Figure PCTCN2022085126-appb-000009
上式中,
Figure PCTCN2022085126-appb-000010
为l-1层到l层的权重矩阵,
Figure PCTCN2022085126-appb-000011
为l-1层到l层的偏置,i表示神经元序号,f l()为l层的非线性激活函数。
可选地,步骤2.2中可视化方式为折线图。
根据本发明提供的具体实施例,本发明公开了以下技术效果:
本发明高效利用5G新兴网络架构下有限的航空发动机故障数据资源,结合计算和存储资源,能够提高航空发动机海量运行数据的存储与传输速度,为航空发动机故障识别提供可靠的依据。
本发明利用机器学习技术在边缘云中构建智能学习模块,通过建立具体的一维卷积神经网络,将预先采集或储存在数据库中的航空发动机齿轮原始振动数据(一维时域振动信号)作为输入,经过层层特征提取,完成故障类型的识别;边缘计算技术在靠近网络接入端对数据进行处理,降低数据传输成本,节省时间、提高效率;一维卷积神经网络(1D-CNN)能够直接对时域信号进行特征挖掘,其采集的一维时域振动信号作为样本空间输入网络,免去了原始信号处理过程,能更加快捷地完成故障类型识别和诊断,在海量航空发动机运行数据的处理上具有潜在的应用价值。
与传统的航空旋转机械故障诊断使用的信号处理与机器学习相结合的方法(通过将采集振动时域信号切片、采用信号处理方法转换为时频图、再采取图像识别或神经网络进行分类诊断)相比,本发明直接对原始一维振动信号进行卷积神经网络操作,过程较为简易,不需要丰富的信号处理专家经验,识别效果也较为理想。
说明书附图
为了更清楚地说明本发明实施例或现有技术中的技术方案,下面将对实施例中所需要使用的附图作简单地介绍,显而易见地,下面描述中的附图仅仅是本发明的一些实施例,对于本领域普通技术人员来讲,在不付出创造性劳动的前提下,还可以根据这些附图获得其他的附图。
图1为本发明基于5G边缘计算和深度学习的航空发动机故障诊断方法的流程图;
图2为航空发动机齿轮故障信号数据图;
图3为一维卷积神经网络结构示意图;
图4为一维卷积神经网络模型训练结果可视化示意图。
具体实施方式
下面将结合本发明实施例中的附图,对本发明实施例中的技术方案进行清楚、完整地描述,显然,所描述的实施例仅仅是本发明一部分实施例,而不是全部的实施例。基于本发明中的实施例,本领域普通技术人员在没有做出创造性劳动前提下所获得的所有其他实施例,都属于本发明保护的范围。
实施例1:
本实施例提供一种基于5G边缘计算和深度学习的航空发动机故障诊断方法,其流程如图1所示,具体包括以下步骤:
步骤1、基于5G新型云边端网络架构的数据采集、预处理及存储。
步骤1.1、数据采集:搭建航空发动机齿轮故障模拟平台,采用边缘计算技术(5G核心技术),在靠近航空发动机齿轮故障模拟平台的边缘网络布置基站,用于数据采集,直接在网络边缘进行数据的处理、传输与储存,避免数据返回核心网络二产生的时延、损失等;通过航空发动机齿轮故障模拟平台安装的加速度传感器采集不同故障类型的齿轮不同位置和方向的振动信号,将振动信号转换为电压信号;基站通过配置包含的时隙数目K来保证每个航空发动机组内的终端设备都满足业务传输的时延约束:
Figure PCTCN2022085126-appb-000012
其中,T I是终端设备初次传输数据的时长;T R是终端设备初次数据传输出错后重新传输数据的时长;T n(1≤n≤N Tot,n∈N +)是终端设备n初次传输数据失败的时刻到下一次重新发送数据时刻之间的时间间隔;N Tot是一个终端设备分组中的总的终端设备数;T Threshold是业务传输的时延约束。
基站按照同一个分组内的终端设备数目为每个终端设备分配独立的初次数据传输资源;同一个分组内的终端设备在初次数据传输失败后,等待基站配置重传资源时刻,再重新传输数据。
包含的时隙数目K为每个分组的大小;时隙数目K设置为:
Figure PCTCN2022085126-appb-000013
其中,T Threshold是业务传输的时延约束,T S是一个数据传输时隙的长度,T I是终端设备初次传输数据的时长;T R是终端设备初次数据传输出错后重新传输数据的时长。
步骤1.2、数据采集系统利用数据库技术,如mySQL、Oracle等来建立航空发动机故障数据库管理系统,实现数据的交互和有效存储;对数据进行预处理和存储。
步骤1.2.1、航空发动机故障数据库管理系统与航空发动机和云端进行数据交互,接受来自航空发动机的数据并,提前为航空发动机缓存数据,向云端上传数据。
步骤1.2.2、将步骤1.1采集得到的原始振动信号对应的电压信号进行缺失值处理、异常值处理和归一化处理,并对不同故障类型的数据进行标记,以保证后续模型训练的效率。缺失值处理为通过缺失值两侧值的平均值补全缺失值;异常值处理为舍弃异常值(其中异常值为仪器误差引起的数据突然短暂性波动);归一化处理采用最大最小值归一化,归一化公式为:
Figure PCTCN2022085126-appb-000014
其中,xmax为样本数据的最大值,xmin为样本数据的最小值,x`为归一化结果,样本数据的数值区间为[0,1]。
步骤2、利用机器学习技术在边缘云中构建机器学习模块,航空发动机故障数据库管理系统存储的历史数据为机器学习模块提供训练样本;机器学习模块利用来自航空发动机故障数据库管理系统的数据,通过一维卷积神经网络模型对航空发动机行为进行预测和推理,对通信和计算资源进行联合优化分配。
步骤2.1、采用Python语言搭建一维卷积神经网络模型;一维卷积神经网络模型包括一个输入层,五个卷积层,五个池化层,一个全连接层和一个输出层(具体网络结构可以根据特定的数据进行调整);一维卷积神经网络模型的输入层分别连接五个卷积层,五个卷积层分别连接五个池化层,五个池化层汇总连接全连接层,全连接层连接输出层;通过一维卷积神经网络模型对振动信号进行特征提取和类型识别,最终输出各故障类别振动信号的概率值作为识别结果;
输入层特征映射组为二维张量,其中每个切片一维数组为一个输入特征映射,输入层的通道数为航空发动机齿轮故障模拟平台安装的加速度传感器数量;将传感器采集的不同故障类型的齿轮不同位置和方向的振动信号分别设置为数据特征,将故障类别设置为数据标签。
卷积层中每两层神经网络之间连接三个神经元,卷积过程中局部区域特征提取的公式为:
Figure PCTCN2022085126-appb-000015
其中,
Figure PCTCN2022085126-appb-000016
表示第i个卷积核在l层的权重,
Figure PCTCN2022085126-appb-000017
表示第i个卷积核在l层的偏置,x l(j)表示第l层的第j个神经元的输入,
Figure PCTCN2022085126-appb-000018
表示第j个神经元在l+1层的输入,
Figure PCTCN2022085126-appb-000019
同时为l层的输出;f()为非线性激活函数,一般选用Relu函数,符号*表示内核与该局部区域的点积;每次卷积运算完成后,将输出特征映射为一维张量。
池化层采取最大值池化方法,选择该池化区域所有神经元的最大活性值作为这个池化区域的表示,池化函数的表达式为:
y d=max x i(i∈R d)
其中,x i为池化区域R d内每个神经元的活性值;池化层对于每个特征映射的划分不宜过大,池化核设置为2×1。
全连接层对卷积层和池化层提取的特征进行非线性组合:
Figure PCTCN2022085126-appb-000020
其中,
Figure PCTCN2022085126-appb-000021
为l-1层到l层的权重矩阵,
Figure PCTCN2022085126-appb-000022
为l-1层到l层的偏置,i表示神经元序号,f l()为l层的非线性激活函数,选用Relu函数。
步骤2.2、一维卷积神经网络模型的训练及结果可视化:将上述处理后的航空发动机振动信号输入待训练的一维卷积神经网络,设置训练集和测试集的比例(一般为4:1,数据量较大时可以适当提高测试集占比)、模型迭代次数(取500次,也可以根据数据量进行适当调整)、单次训练送入数据批量、训练批次和网络参数(32个样本,或16的整数倍),并实时监测一维卷积神经网络模型的识别准确率和损失函数值的变化;以折线图方式输出一维卷积神经网络模型的识别效果。
步骤2.3、对通信和计算资源进行联合优化分配时采用的资源联合优化分配实现模型如下:
min η
s.t.P Suc(D>D threshold)≤ε
其中,ε为超高可靠低延时通信(URLLC)业务的数据包最大容忍错误概 率,D为数据包传输经历的实际时延,P Suc(D>D threshold)≤ε为概率时延约束,η为超高可靠低延时通信(URLLC)业务的资源占比;当URLLC业务和增强移动宽带(eMBB)业务以资源预留模式复用时,URLLC的资源占比η越大,则URLLC业务的服务质量越好,但对eMBB业务的影响也越大,因此要在保证URLLC的高服务质量(QoS)约束下,最小化η,使其对eMBB业务相应最小。
步骤3、航空发动机齿轮故障模拟平台和航空发动机故障数据库管理系统智能化自我管理:在航空发动机齿轮故障模拟平台内部设计决策中心,决策中心接受来自机器学习模块的输出,利用博弈论等工具对机器学习模块机器学习的结果进行分析与决策,实现计算卸载、边云协同、资源优化分配等功能;决策中心同时对航空发动机故障数据库管理系统进行智能管理,指导航空发动机故障数据库管理系统进行提前缓存。
实施例2:
在实施例1的基础上,如图2所示,为航空发动机齿轮故障模拟平台采集的同一传感器布置下五种不同故障类型(正常齿轮(a)、断齿(b)、缺齿(c)、齿面磨损(d)、齿根裂纹(e),可根据具体情况增添和删减)的振动信号数据(横轴为采样时间,纵轴为加速度传感器采集的振幅信号转换为电压值),该数据亦可为通过航空发动机在真实运行情况下采集的实时状态数据。数据获取时设定的采样频率和传感器位置布置可以根据实际情况确定,必要时为了提高模型训练准确度,可增添声学传感器等其他类型的传感器。获取原始数据后,对其进行数据预处理,通过平均值补全缺失值,舍弃异常值,最终进行归一化操作。根据不同的故障类型,本实施例需要将数据进行类别标定,具体标签设置如下表1:
表1 齿轮数据集的标签
Figure PCTCN2022085126-appb-000023
图3为搭建的一维卷积神经网络结构示意图,本实施例提出的一维卷积神经网络模型的具体结构由五个卷积层、五个池化层、一个全连接层和一个Softmax输出层组成。经过第一层卷积层后,信号被转换成一组特征映射,然后通过最大值池化对其进行下采样。在这些操作重复4次后,将最后一个池化层的特性连接到全连接层,然后通过Relu函数激活全连接层,传递到Softmax层,最终得到每个分类的概率值,其中概率最大的类别会被视为识别结果。每次池化操作后需要添加一个Dropout层,舍弃率为20%,即舍弃20%的训练参数,提高模型训练速度,防止过拟合。网络具体参数(卷积核滑动步长等)设置如下表2所示。
表2 一维卷积神经网络的详细参数
Figure PCTCN2022085126-appb-000024
神经网络模型搭建完成后,需将预处理完成的数据按照训练集/测试集为4/1的比例输入到神经网络,设置模型迭代次数为100,每次训练批量为32,最终通过matplotlib绘制准确率和损失函数值实时变化情况(如图4所示),可进一步绘制识别结果的混淆矩阵,观察模型具体识别情况。根据模型训练效果,可逐步调整该一维卷积神经网络模型的参数,优化训练效果。
与传统的航空旋转机械故障诊断使用的信号处理与机器学习相结合的方法(通过将采集振动时域信号切片、采用信号处理方法转换为时频图、再采取图像识别或神经网络进行分类诊断)相比,本实施例直接对原始一维振动信号 进行卷积神经网络操作,过程较为简易,不需要丰富的信号处理专家经验,识别效果也较为理想,最低识别率能达到78.97%。如下表3所示为本实施例的方法与传统机器学习方法的比较结果表。
表3 各机器学习方法识别效果
Figure PCTCN2022085126-appb-000025
相比使用普通的前馈神经网络63.90%的识别准确率提高了15.07%,相较于使用支持向量机63.08%的准确率提高了15.89%,并且该方法对齿轮故障诊断的均方根误差更低。进一步针对特定数据进行参数优化,能够较为准确的识别航空发动机故障。
本说明书中各个实施例采用递进的方式描述,每个实施例重点说明的都是与其他实施例的不同之处,各个实施例之间相同相似部分互相参见即可。
本文中应用了具体个例对本发明的原理及实施方式进行了阐述,以上实施例的说明只是用于帮助理解本发明的方法及其核心思想;同时,对于本领域的一般技术人员,依据本发明的思想,在具体实施方式及应用范围上均会有改变之处。综上所述,本说明书内容不应理解为对本发明的限制。

Claims (9)

  1. 一种基于5G边缘计算和深度学习的航空发动机故障诊断方法,其特征在于,包括以下步骤:
    步骤1、基于5G新型云边端网络架构的数据采集、预处理及存储;
    步骤1.1、数据采集:搭建航空发动机齿轮故障模拟平台,采用边缘计算技术,在靠近所述航空发动机齿轮故障模拟平台的边缘网络布置基站;通过所述航空发动机齿轮故障模拟平台安装的加速度传感器采集齿轮不同位置和方向的振动信号,将振动信号转换为电压信号;其中基站通过配置包含的时隙数目K来保证每个航空发动机组内的终端设备都满足业务传输的时延约束:
    Figure PCTCN2022085126-appb-100001
    其中,T I是终端设备初次传输数据的时长;T R是终端设备初次数据传输出错后重新传输数据的时长;T n(1≤n≤N Tot,n∈N +)是终端设备n初次传输数据失败的时刻到下一次重新发送数据时刻之间的时间间隔;N Tot是一个终端设备分组中的总的终端设备数;T Threshold是业务传输的时延约束;
    基站按照同一个分组内的终端设备数目为每个终端设备分配独立的初次数据传输资源;同一个分组内的终端设备在初次数据传输失败后,等待基站配置重传资源时刻,再重新传输数据;
    包含的时隙数目K为每个分组的大小;时隙数目K设置为:
    Figure PCTCN2022085126-appb-100002
    其中,T Threshold是业务传输的时延约束,T S是一个数据传输时隙的长度,T I是终端设备初次传输数据的时长;T R是终端设备初次数据传输出错后重新传输数据的时长;
    步骤1.2、建立航空发动机故障数据库管理系统;对数据进行预处理和存储;
    步骤2、在边缘云中构建机器学习模块,所述航空发动机故障数据库管理系统存储的历史数据为机器学习模块提供训练样本;所述机器学习模块利用来自所述航空发动机故障数据库管理系统的数据,通过一维卷积神经网络模型对航空发动机行为进行预测和推理,对通信和计算资源进行联合优化分配;
    步骤2.1、搭建一维卷积神经网络模型;一维卷积神经网络模型包括一个输入层,五个卷积层,五个池化层,一个全连接层和一个输出层;通过所述一 维卷积神经网络模型对振动信号进行特征提取和类型识别,最终输出各故障类别振动信号的概率值作为识别结果;
    步骤2.2、一维卷积神经网络模型的训练及结果可视化:将处理后的航空发动机振动信号输入待训练的一维卷积神经网络,设置训练集和测试集的比例、模型迭代次数、单次训练送入数据批量、训练批次和网络参数,并实时监测一维卷积神经网络模型的识别准确率和损失函数值的变化;以可视化方式输出一维卷积神经网络模型的识别效果;
    步骤2.3、对通信和计算资源进行联合优化分配时采用的资源联合优化分配实现模型如下:
    minη
    s.t.P Suc(D>D threshold)≤ε
    其中,ε为超高可靠低延时通信业务的数据包最大容忍错误概率,D为数据包传输经历的实际时延,P Suc(D>D threshold)≤ε为概率时延约束,η为超高可靠低延时通信业务的资源占比;
    步骤3、航空发动机齿轮故障模拟平台和航空发动机故障数据库管理系统智能化自我管理:在所述航空发动机齿轮故障模拟平台内部设计决策中心,决策中心接受来自所述机器学习模块的输出,对机器学习模块机器学习的结果进行分析与决策;决策中心同时对所述航空发动机故障数据库管理系统进行管理,指导所述航空发动机故障数据库管理系统进行提前缓存。
  2. 根据权利要求1所述基于5G边缘计算和深度学习的航空发动机故障诊断方法,其特征在于,步骤1.2具体包括以下步骤:
    步骤1.2.1、所述航空发动机故障数据库管理系统与航空发动机和云端进行数据交互,接受来自航空发动机的数据并缓存数据,向云端上传数据;
    步骤1.2.2、将步骤1.1采集得到的原始振动信号对应的电压信号进行缺失值处理、异常值处理和归一化处理,并对不同故障类型的数据进行标记。
  3. 根据权利要求2所述基于5G边缘计算和深度学习的航空发动机故障诊断方法,其特征在于,步骤1.2.2中缺失值处理为通过缺失值两侧值的平均值补全缺失值;异常值处理为舍弃异常值;归一化处理采用最大最小值归一化,归一化公式为:
    Figure PCTCN2022085126-appb-100003
    其中,xmax为样本数据的最大值,xmin为样本数据的最小值,x`为归一化结果,样本数据的数值区间为[0,1]。
  4. 根据权利要求1所述基于5G边缘计算和深度学习的航空发动机故障诊断方法,其特征在于,步骤2.1中一维卷积神经网络模型的输入层分别连接五个卷积层,五个卷积层分别连接五个池化层,五个池化层汇总连接全连接层,全连接层连接输出层。
  5. 根据权利要求1所述基于5G边缘计算和深度学习的航空发动机故障诊断方法,其特征在于,步骤2.1中一维卷积神经网络模型的输入层特征映射组为二维张量,其中每个切片一维数组为一个输入特征映射,输入层的通道数为航空发动机齿轮故障模拟平台安装的加速度传感器数量;将传感器采集的不同故障类型的齿轮不同位置和方向的振动信号分别设置为数据特征,将故障类别设置为数据标签。
  6. 根据权利要求1所述基于5G边缘计算和深度学习的航空发动机故障诊断方法,其特征在于,步骤2.1中一维卷积神经网络模型的卷积层中每两层神经网络之间连接三个神经元,卷积过程中局部区域特征提取的公式为:
    Figure PCTCN2022085126-appb-100004
    其中,
    Figure PCTCN2022085126-appb-100005
    表示第i个卷积核在l层的权重,
    Figure PCTCN2022085126-appb-100006
    表示第i个卷积核在l层的偏置,x l(j)表示第l层的第j个神经元的输入,
    Figure PCTCN2022085126-appb-100007
    表示第j个神经元在l+1层的输入,
    Figure PCTCN2022085126-appb-100008
    同时为l层的输出;f()为非线性激活函数,符号*表示内核与该局部区域的点积;每次卷积运算完成后,将输出特征映射为一维张量。
  7. 根据权利要求1所述基于5G边缘计算和深度学习的航空发动机故障诊断方法,其特征在于,步骤2.1中一维卷积神经网络模型的池化层采取最大值池化方法,选择该池化区域所有神经元的最大活性值作为这个池化区域的表示,池化函数的表达式为:
    y d=max x i(i∈R d)
    其中,x i为池化区域R d内每个神经元的活性值。
  8. 根据权利要求1所述基于5G边缘计算和深度学习的航空发动机故障诊断方法,其特征在于,步骤2.1中一维卷积神经网络模型的全连接层对卷积 层和池化层提取的特征进行非线性组合:
    Figure PCTCN2022085126-appb-100009
    其中,
    Figure PCTCN2022085126-appb-100010
    为l-1层到l层的权重矩阵,
    Figure PCTCN2022085126-appb-100011
    为l-1层到l层的偏置,i表示神经元序号,f l()为l层的非线性激活函数。
  9. 根据权利要求1所述基于5G边缘计算和深度学习的航空发动机故障诊断方法,其特征在于,步骤2.2中可视化方式为折线图。
PCT/CN2022/085126 2021-04-27 2022-04-02 基于5g边缘计算和深度学习的航空发动机故障诊断方法 WO2022228049A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202110457105.XA CN112989712B (zh) 2021-04-27 2021-04-27 一种基于5g边缘计算和深度学习的航空发动机故障诊断方法
CN202110457105.X 2021-04-27

Publications (1)

Publication Number Publication Date
WO2022228049A1 true WO2022228049A1 (zh) 2022-11-03

Family

ID=76340250

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2022/085126 WO2022228049A1 (zh) 2021-04-27 2022-04-02 基于5g边缘计算和深度学习的航空发动机故障诊断方法

Country Status (2)

Country Link
CN (1) CN112989712B (zh)
WO (1) WO2022228049A1 (zh)

Cited By (11)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114893390A (zh) * 2022-07-15 2022-08-12 安徽云磬科技产业发展有限公司 一种基于注意力与集成学习机制的泵机设备故障检测方法
CN115600138A (zh) * 2022-12-13 2023-01-13 四川大学(Cn) 基于动态图残差卷积的流体动压密封环磨损故障检测方法
CN116467674A (zh) * 2023-04-10 2023-07-21 国网辽宁省电力有限公司抚顺供电公司 一种配电网智能故障处理融合更新系统及其方法
CN116502075A (zh) * 2023-06-28 2023-07-28 吉林大学 一种多模态水下自主航行器状态检测方法及系统
CN116728291A (zh) * 2023-08-16 2023-09-12 湖南大学 基于边缘计算的机器人打磨系统状态监测方法和装置
CN116953660A (zh) * 2023-09-18 2023-10-27 中国科学院精密测量科学与技术创新研究院 一种全高层大气风温密探测激光雷达云边协同方法
CN117009791A (zh) * 2023-09-28 2023-11-07 太仓点石航空动力有限公司 一种航空发动机的故障识别方法及系统
CN117148775A (zh) * 2023-10-31 2023-12-01 中国电建集团山东电力管道工程有限公司 管道生产过程远程监控方法、系统、设备及介质
CN117572779A (zh) * 2024-01-15 2024-02-20 北京航空航天大学杭州创新研究院 一种桨叶损伤下的电动航空发动机控制方法
CN117834724A (zh) * 2024-03-04 2024-04-05 中科软股教育科技(北京)股份有限公司 一种基于大数据分析的视频学习资源管理系统
CN117851954A (zh) * 2024-03-06 2024-04-09 大连海泰轴承制造有限公司 基于数据分析的轴承加工设备运行质量检测系统和方法

Families Citing this family (13)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112989712B (zh) * 2021-04-27 2021-08-03 浙大城市学院 一种基于5g边缘计算和深度学习的航空发动机故障诊断方法
CN113255584B (zh) * 2021-06-22 2021-10-19 德明通讯(上海)股份有限公司 一种基于边缘计算的故障诊断与监测系统
CN113469060A (zh) * 2021-07-02 2021-10-01 浙大城市学院 多传感器融合卷积神经网络航空发动机轴承故障诊断方法
CN113758709A (zh) * 2021-09-30 2021-12-07 河南科技大学 结合边缘计算和深度学习的滚动轴承故障诊断方法及系统
CN114062511A (zh) * 2021-10-24 2022-02-18 北京化工大学 一种基于单传感器的航空发动机早期损伤声发射智能识别方法
CN114330413A (zh) * 2021-11-25 2022-04-12 中车永济电机有限公司 牵引电机轴承的故障类型辨识及定位方法
CN114564882A (zh) * 2022-01-29 2022-05-31 内蒙古工业大学 基于离散事件的边缘深度学习模拟器的构建与应用
CN114781522A (zh) * 2022-04-26 2022-07-22 苏州今科慧邦科技有限公司 一种基于改进卷积神经网络的航空器附件在线检测系统及方法
CN114970645B (zh) * 2022-07-27 2023-04-07 中国工业互联网研究院 基于5g边云协同的民用航空发动机故障诊断系统及方法
CN116306095A (zh) * 2023-01-16 2023-06-23 中国电建集团北京勘测设计研究院有限公司 基于边缘计算的抽水蓄能机组故障诊断系统及方法
CN116718373B (zh) * 2023-06-13 2024-01-05 长江勘测规划设计研究有限责任公司 一种齿轮齿条驱动机构故障特征信号识别方法及装置
CN116484268B (zh) * 2023-06-21 2023-09-05 西安黑石智能科技有限公司 基于机器学习的智能化工业设备故障诊断系统
CN117216659A (zh) * 2023-09-12 2023-12-12 暨南大学 基于单颗粒气溶胶质谱的大气颗粒物来源解析方法及系统

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20200233397A1 (en) * 2019-01-23 2020-07-23 New York University System, method and computer-accessible medium for machine condition monitoring
US20200272139A1 (en) * 2019-02-21 2020-08-27 Abb Schweiz Ag Method and System for Data Driven Machine Diagnostics
US20200327371A1 (en) * 2019-04-09 2020-10-15 FogHorn Systems, Inc. Intelligent Edge Computing Platform with Machine Learning Capability
CN111830408A (zh) * 2020-06-23 2020-10-27 朗斯顿科技(北京)有限公司 一种基于边缘计算和深度学习的电机故障诊断系统及方法
US20200380391A1 (en) * 2019-05-29 2020-12-03 Caci, Inc. - Federal Methods and systems for predicting electromechanical device failure
CN112101532A (zh) * 2020-11-18 2020-12-18 天津开发区精诺瀚海数据科技有限公司 一种基于边云协同的自适应多模型驱动设备故障诊断方法
CN112989712A (zh) * 2021-04-27 2021-06-18 浙大城市学院 一种基于5g边缘计算和深度学习的航空发动机故障诊断方法

Family Cites Families (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20170178311A1 (en) * 2015-12-20 2017-06-22 Prophecy Sensors, Llc Machine fault detection based on a combination of sound capture and on spot feedback
CN108344574B (zh) * 2018-04-28 2019-09-10 湖南科技大学 一种基于深度联合适配网络的风电机组轴承故障诊断方法
CN110221558B (zh) * 2019-06-05 2020-09-01 镇江四联机电科技有限公司 一种基于边缘计算技术的电液伺服阀在线故障诊断网关
CN112067916A (zh) * 2019-09-20 2020-12-11 武汉理工大学 基于深度学习的时间序列数据智能故障诊断方法
CN111259532B (zh) * 2020-01-13 2022-05-27 西北工业大学 基于3dcnn-jtfa的航空发动机控制系统传感器的故障诊断方法
CN111597760B (zh) * 2020-05-18 2022-07-22 哈尔滨工业大学(威海) 一种实现小样本条件下获取气路参数偏差值的方法
CN112083244B (zh) * 2020-08-30 2022-10-28 西南电子技术研究所(中国电子科技集团公司第十研究所) 综合化航空电子设备故障智能诊断系统
CN112101767B (zh) * 2020-09-09 2023-12-26 中国石油大学(北京) 一种设备运行状态边云融合诊断方法及系统

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20200233397A1 (en) * 2019-01-23 2020-07-23 New York University System, method and computer-accessible medium for machine condition monitoring
US20200272139A1 (en) * 2019-02-21 2020-08-27 Abb Schweiz Ag Method and System for Data Driven Machine Diagnostics
US20200327371A1 (en) * 2019-04-09 2020-10-15 FogHorn Systems, Inc. Intelligent Edge Computing Platform with Machine Learning Capability
US20200380391A1 (en) * 2019-05-29 2020-12-03 Caci, Inc. - Federal Methods and systems for predicting electromechanical device failure
CN111830408A (zh) * 2020-06-23 2020-10-27 朗斯顿科技(北京)有限公司 一种基于边缘计算和深度学习的电机故障诊断系统及方法
CN112101532A (zh) * 2020-11-18 2020-12-18 天津开发区精诺瀚海数据科技有限公司 一种基于边云协同的自适应多模型驱动设备故障诊断方法
CN112989712A (zh) * 2021-04-27 2021-06-18 浙大城市学院 一种基于5g边缘计算和深度学习的航空发动机故障诊断方法

Cited By (21)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114893390A (zh) * 2022-07-15 2022-08-12 安徽云磬科技产业发展有限公司 一种基于注意力与集成学习机制的泵机设备故障检测方法
CN115600138A (zh) * 2022-12-13 2023-01-13 四川大学(Cn) 基于动态图残差卷积的流体动压密封环磨损故障检测方法
CN115600138B (zh) * 2022-12-13 2023-06-20 四川大学 基于动态图残差卷积的流体动压密封环磨损故障检测方法
CN116467674A (zh) * 2023-04-10 2023-07-21 国网辽宁省电力有限公司抚顺供电公司 一种配电网智能故障处理融合更新系统及其方法
CN116467674B (zh) * 2023-04-10 2024-05-17 国网辽宁省电力有限公司抚顺供电公司 一种配电网智能故障处理融合更新系统及其方法
CN116502075A (zh) * 2023-06-28 2023-07-28 吉林大学 一种多模态水下自主航行器状态检测方法及系统
CN116502075B (zh) * 2023-06-28 2023-09-12 吉林大学 一种多模态水下自主航行器状态检测方法及系统
CN116728291A (zh) * 2023-08-16 2023-09-12 湖南大学 基于边缘计算的机器人打磨系统状态监测方法和装置
CN116728291B (zh) * 2023-08-16 2023-10-31 湖南大学 基于边缘计算的机器人打磨系统状态监测方法和装置
CN116953660A (zh) * 2023-09-18 2023-10-27 中国科学院精密测量科学与技术创新研究院 一种全高层大气风温密探测激光雷达云边协同方法
CN116953660B (zh) * 2023-09-18 2024-01-05 中国科学院精密测量科学与技术创新研究院 一种全高层大气风温密探测激光雷达云边协同方法
CN117009791A (zh) * 2023-09-28 2023-11-07 太仓点石航空动力有限公司 一种航空发动机的故障识别方法及系统
CN117009791B (zh) * 2023-09-28 2023-12-12 太仓点石航空动力有限公司 一种航空发动机的故障识别方法及系统
CN117148775A (zh) * 2023-10-31 2023-12-01 中国电建集团山东电力管道工程有限公司 管道生产过程远程监控方法、系统、设备及介质
CN117148775B (zh) * 2023-10-31 2024-01-23 中国电建集团山东电力管道工程有限公司 管道生产过程远程监控方法、系统、设备及介质
CN117572779A (zh) * 2024-01-15 2024-02-20 北京航空航天大学杭州创新研究院 一种桨叶损伤下的电动航空发动机控制方法
CN117572779B (zh) * 2024-01-15 2024-03-26 北京航空航天大学杭州创新研究院 一种桨叶损伤下的电动航空发动机控制方法
CN117834724A (zh) * 2024-03-04 2024-04-05 中科软股教育科技(北京)股份有限公司 一种基于大数据分析的视频学习资源管理系统
CN117834724B (zh) * 2024-03-04 2024-04-30 中科软股教育科技(北京)股份有限公司 一种基于大数据分析的视频学习资源管理系统
CN117851954A (zh) * 2024-03-06 2024-04-09 大连海泰轴承制造有限公司 基于数据分析的轴承加工设备运行质量检测系统和方法
CN117851954B (zh) * 2024-03-06 2024-05-24 大连海泰轴承制造有限公司 基于数据分析的轴承加工设备运行质量检测系统和方法

Also Published As

Publication number Publication date
CN112989712B (zh) 2021-08-03
CN112989712A (zh) 2021-06-18

Similar Documents

Publication Publication Date Title
WO2022228049A1 (zh) 基于5g边缘计算和深度学习的航空发动机故障诊断方法
CN111369563B (zh) 一种基于金字塔空洞卷积网络的语义分割方法
Zhang et al. Data-driven methods for predictive maintenance of industrial equipment: A survey
CN111830408B (zh) 一种基于边缘计算和深度学习的电机故障诊断系统及方法
CN110943857B (zh) 基于卷积神经网络的电力通信网故障分析及定位方法
CN110262463B (zh) 一种基于深度学习的轨道交通站台门故障诊断系统
Pandhare et al. Convolutional neural network based rolling-element bearing fault diagnosis for naturally occurring and progressing defects using time-frequency domain features
WO2023123593A1 (zh) 基于变分模态分解和残差网络的航空轴承故障诊断方法
CN111431986B (zh) 基于5g和ai云边协同的工业智能质检系统
CN107179503A (zh) 基于随机森林的风电机组故障智能诊断预警的方法
CN111562105B (zh) 一种基于小波包分解和卷积神经网络的风电机组齿轮箱故障诊断方法
CN109946080B (zh) 一种基于嵌入式循环网络的机械设备健康状态识别方法
CN107908175A (zh) 一种电力系统现场智能化运维系统
CN110647830A (zh) 基于卷积神经网络和高斯混合模型的轴承故障诊断方法
Xu et al. Fault diagnosis of rolling bearing based on online transfer convolutional neural network
CN102521604A (zh) 一种基于巡检系统的设备性能退化评估装置及方法
WO2019178930A1 (zh) 一种机械设备故障诊断方法
CN114004135A (zh) 基于Transformer神经网络的农机轴承故障类型诊断方法及系统
CN114186617B (zh) 一种基于分布式深度学习的机械故障诊断方法
Chen et al. Industrial edge intelligence: Federated-meta learning framework for few-shot fault diagnosis
CN107016440B (zh) 机械传动故障的多分辨率深度神经网络智能诊断方法
Ning et al. An intelligent device fault diagnosis method in industrial internet of things
Zhu et al. Fault diagnosis of wheelset bearings using deep bidirectional long short-term memory network
US20240185040A1 (en) Aero-engine fault diagnosis method based on 5g edge computing and deep learning
CN115623004A (zh) 一种基于区块链的轨道交通设备管理系统及方法

Legal Events

Date Code Title Description
WWE Wipo information: entry into national phase

Ref document number: 17795693

Country of ref document: US

121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 22794521

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 22794521

Country of ref document: EP

Kind code of ref document: A1