Granted Patent B1

US 12,292,369 · App. 18/750,844 · Granted May 6, 2025

Inventors: Zhenghua Qian (Nanjing, CN); Guan Wang (Nanjing, CN); Xianwei Wu (Nanjing, CN); Zhi Qian (Nanjing, CN); Peng Li (Nanjing, CN)

Assignees: Nanjing University of Aeronautics and Astronautics (NUAA); Nanjing University of Aeronautics and Astronautics Shenzhen Research Institute

G01N15/0656G01N33/2858G06N20/00

Abstract

Claims (37)

1. An induction type oil debris signal monitoring and identification method, which is characterized by including:

detecting electromagnetic changes in oil, and outputting a monitoring signal after signal processing;

extracting and analyzing an amplitude from the monitoring signal;

wherein if the amplitude is a bubble signal, it will not be included in the quantity of debris;

wherein analyzing, by a bubble and debris classification and identification network module, if the amplitude is a bubble signal comprises basing the bubble and debris classification and identification network module on a back propagation (BP) neural network, and training the bubble and debris classification and identification network module by inputting an oil bubble signal and a debris signal to achieve automatic separation differentiation of the bubbles and the debris;

wherein if the amplitude is a debris signal, then analyzing the monitoring signal with a two-level debris signal analysis;

during the two-level debris signal analysis, first identifying the quantity of the debris in the monitoring signal;

wherein if it is identified to be a single quantity of debris, then identifying the size of the single quantity of debris;

wherein if it is identified to be a double quantity of debris, then identifying the respective sizes of the double quantity of debris; and

wherein if it is identified to be a triple quantity of debris, then identifying the respective sizes of the triple quantity of debris; and

recording the quantity or quantities and size or sizes of the debris.

2. An induction type oil debris signal monitoring and identification method according to claim 1 , further comprising:

detecting, by an induction type oil debris sensor, the electromagnetic changes in the oil, and outputting the monitoring signal after processing through a signal processing circuit; and

analyzing, by a two-level debris signal analysis network module, the monitoring signal and determining the size and quantity of the debris.

3. The induction type oil debris signal monitoring and identification method according to claim 2 , further comprising:

collecting, by the bubble and debris classification and identification network module, bubble and debris training signals comprising:

collecting bubble signals including: using a pinhole to inject bubbles of different sizes into the oil, and collecting bubble signals to ensure diversity of the collected signals;

collecting debris training signals is as follows: randomly using single quantity debris, double quantity debris, or triple quantity debris in the oil, and collecting debris training signals to ensure the diversity of the collected signals; and

setting labels for the bubble signal and debris training signals respectively to obtain the first collected data; and

wherein the-bubble and debris classification and identification model training includes:

using a BP algorithm toolbox for model training; and

randomly selecting training data from ⅓ of all the first collected data and dividing into a training set, a verification set and a test set in a ratio of 7:2:1; wherein a training learning rate is a dynamic learning rate; and during the training process, selecting different numbers of hidden layers and nodes to repeat the training, and determining a best model based on a comprehensive accuracy rate and network size.

4. The induction type oil debris signal monitoring and identification method according to claim 2 , further comprising:

collecting, by the two level debris signal analysis module, debris training signals including:

collecting the signals multiple times is as follows: randomly using single-particle, double-particle, and triple-particle debris in the oil, and collecting debris training signals to obtain a second collected data; and

training the debris quantity identification model including:

using a BP algorithm toolbox for model training; and

randomly selecting training data from ⅓ of all the second collected data and dividing into a training set, a verification set and a test set in a ratio of 7:2:1, wherein a training learning rate is a dynamic learning rate; and selecting an optimal number of hidden layers and nodes through multiple tests, and selecting best results to determine a best model based on a comprehensive accuracy rate and network size.

5. The induction type oil debris signal monitoring and identification method according to claim 4 , characterized in that when the debris training signal is collected, intercepting 100 sampling points from the signal, and locating interception points in the middle of peak and valley time points of the signal.

6. The induction type oil debris signal monitoring and identification method according to claim 2 , wherein

the induction type oil debris sensor comprises an enameled wire wound on a ceramic frame, forming two inverse parallel excitation coils and an electromagnetic induction coil; and

outputting an electrical signal from the electromagnetic induction coil, and amplifying, self-multiplying, demodulating and filtering the signal through the signal processing circuit, and outputting the monitoring signal.

7. The induction type oil debris signal monitoring and identification method according to claim 2 , wherein

basing the two-level debris signal analysis network module on a back propagation neural network, and

constructing the two-level debris signal analysis network module using a two-level network analysis method; wherein

using a first-level network in the two-level debris signal analysis network module to identify the quantity of the debris in the monitoring signal, and using a second-level network in the two-level debris signal analysis network module to identify the size of the debris in the monitoring signal.

8. The induction type oil debris signal monitoring and identification method according to claim 7 , wherein the second-level network further includes using a single debris size identification network, a double debris size identification network and a triple debris size identification network.

Assignments (2)

Priority Claims (1)

References Cited (12)