Aero-engine fault diagnosis method based on fifth-generation telecommunication technology standard distributed computing framework and deep learning

An aero-engine fault diagnosis method based on a fifth-generation telecommunication technology standard distributed computing framework and deep learning, comprising following steps:

• step 1:
• performing data acquisition, preprocessing, and storage based on a fifth-generation telecommunication technology standard distributed computing framework terminal network architecture, wherein step 1 comprises following steps:
• step 1.1:
• performing data acquisition, which comprises:
• building an aero-engine gear fault simulation platform, and arranging, by using an edge computing technology, a base station in an edge network close to the aero-engine gear fault simulation platform, and acquiring vibration signals of gears in different positions and directions by acceleration sensors mounted on the aero-engine gear fault simulation platform, and converting the vibration signals into voltage signals, wherein the base station, by configuring a timeslot number K contained therein, ensures that all terminal devices in each aero-engine meet a delay constraint of service transmission:
• TI+TR+max{TI, . . . ,TNTot}≤TThreshold;
• wherein TI is a duration during which each terminal device transmits data for a first time, TR is a duration during which each terminal device retransmits the data after failing to transmit the data for the first time, Tn(1≤n≤NTot, n∈N+) is a time interval between a moment at which a terminal device n fails to transmit the data for the first time and a moment of retransmitting the data next time, NTot is a total number of the terminal devices in a terminal device group, and TThreshold is a delay constraint of service transmission;
• the base station allocates independent initial data transmission resources to each terminal device according to a number of the terminal devices in a same group, and after failing to transmit the data for the first time, the terminal devices in the same group retransmit the data after the base station configures retransmission resources;
• the contained timeslot number K is the number of timeslots of each group the timeslot number K of each group is set to:
• K=└[TThreshold−(TI+TR)]/TS┘
• wherein TThreshold is the delay constraint of the service transmission, TS is a length of a data transmission timeslot, and TI is the duration during which each terminal device transmits the data for the first time, and TR is the duration during which each terminal device retransmits the data after failing to transmit the data for the first time;
• and
• step 1.2:
• establishing an aero-engine fault database management system, and preprocessing and storing the data;
• step 2:
• constructing a machine learning module in an edge cloud, wherein historical data stored in the aero-engine fault database management system is used as training samples of the machine learning module, and the machine learning module predicts and infers a behavior of an aero-engine through a one-dimensional convolutional neural network (1D-CNN) model by using the data from the aero-engine fault database management system, and performs joint optimization allocation on communication and computing resources, wherein step 2 comprises following steps;
• step 2.1:
• building the 1D-CNN model, wherein the 1D-CNN model comprises one input layer, five convolutional layers, five pooling layers, one fully-connected layer, and one output layer, and performing feature extraction and type recognition on the vibration signals by using the 1D-CNN model, and outputting probability values of the vibration signals under various fault types, as a recognition result;
• step 2.2:
• training the 1D-CNN model and visualizing a result of the 1D-CNN model, which comprises:
• inputting the processed vibration signals of the aero-engine into a to-be-trained 1D-CNN model, setting a ratio of training sets to test sets, a number of iterations of the model, a batch size of data sent into the model for a single-time training, a number of training batches, and network parameters, and monitoring recognition accuracy of the 1D-CNN model and a change in a loss function value in real time;
• and outputting a recognition result of the 1D-CNN model;
• and
• step 2.3:
• using a following model for implementing resource joint optimization allocation used when joint optimization allocation is performed on communication and computing resources:
• min
• η
• s
• .
• t
• .
• P
• Suc
• (
• D
• >
• D
• threshold
• )
• ≤
• ε
• wherein ε is a tolerable maximum value of error probability for a data packet of ultra-reliable and low-latency communication (URLLC) service, D is an actual delay of data packet transmission, PSuc (D>Dthreshold)≤ε is a probabilistic delay constraint, and η is a resource ratio of the URLLC service;
• and
• step 3:
• performing self-management of the aero-engine gear fault simulation platform and the aero-engine fault database management system, which comprises:
• providing a decision center inside the aero-engine gear fault simulation platform, wherein the decision center receives an output from the machine learning module, and analyzes and makes decisions on a machine learning result of the machine learning module;
• and the decision center also manages the aero-engine fault database management system, and instructs the aero-engine fault database management system to cache in advance.