Search

CN-122016323-A - Engine part fault prediction system based on multi-inertial sensor fusion

CN122016323ACN 122016323 ACN122016323 ACN 122016323ACN-122016323-A

Abstract

The invention discloses an engine part fault prediction system based on multi-inertial sensor fusion, and particularly relates to the technical field of engine state monitoring and fault prediction, comprising the steps of collecting multi-dimensional inertial motion data of key parts; the method comprises the steps of receiving multidimensional inertial motion data, dynamically adjusting a sensor array data acquisition strategy based on real-time working condition parameters of an engine, generating a fusion feature vector, outputting a state feature vector of the health state of the engine part, and deducing a future health state degradation track of the engine part through probability simulation. According to the invention, the engine component fault prediction system fused by the multiple inertial sensors is constructed through the multiple inertial sensor acquisition module, the self-adaptive synchronous acquisition module, the space-time characteristic fusion module, the component state agent model module and the fault evolution reasoning module, so that the problems of insufficient comprehensiveness and synchronism of data acquisition, insufficient effectiveness of characteristic fusion, insufficient refinement degree of state evaluation and the like are solved.

Inventors

  • JIANG XIAOYI

Assignees

  • 浙江长江机械有限公司

Dates

Publication Date
20260512
Application Date
20260205

Claims (8)

  1. 1. An engine component fault prediction system based on multi-inertial sensor fusion, comprising: the multi-inertial sensor acquisition module is arranged on a key component measuring point of the engine and used for acquiring multi-dimensional inertial motion data of the key component; The self-adaptive synchronous acquisition module is connected with the multi-inertial sensor array module and is used for receiving multi-dimensional inertial motion data and dynamically adjusting a sensor array data acquisition strategy based on real-time working condition parameters of an engine; The space-time feature fusion module is connected with the self-adaptive synchronous acquisition module and used for preprocessing multidimensional inertial motion data to generate a fusion feature vector containing a physical field coupling feature vector and a transfer path feature vector; The component state agent model module is connected with the space-time feature fusion module and the self-adaptive synchronous acquisition module, receives the fusion feature vector and working condition parameters of the engine, and outputs a state feature vector of the health state of the engine component, wherein the state feature vector comprises a health index, potential fault probability and degradation trend parameters; And the fault evolution reasoning module is connected with the component state agent model and is used for receiving the state characteristic vector and deducing the future health state degradation track of the engine component through probability simulation so as to predict the occurrence probability of the fault.
  2. 2. The multiple inertial sensor fusion based engine component failure prediction system of claim 1, wherein the multiple inertial sensor acquisition module comprises: the measuring point arrangement of the multi-inertial sensor array is based on key components of engine structure analysis and fault statistics data screening, the key components comprise a rotating component, a transmission component, a reciprocating component and a fixed key component, 3 to 6 measuring points are arranged on each key component, and a component movement core area and a force transmission path node are covered.
  3. 3. The multiple inertial sensor fusion based engine component failure prediction system of claim 2, wherein the multiple inertial sensor acquisition module, sensor profiling and mounting, comprises: The MEMS triaxial inertial sensor integrating an accelerometer and a gyroscope is selected, a static or low-vibration part is fixed by high-temperature epoxy glue, the thickness of a glue layer is controlled to be 0.1-0.3 mm, the curing time is longer than 24 hours, and the temperature of a curing environment is controlled to be 20 ℃ plus or minus 5 ℃; the dynamic or high vibration component is fixed by M3 screw thread and is matched with a locknut for locking, and the posture of the sensor is adjusted by a high-precision level meter during installation, so that the movement direction of the accelerometer shaft and the component is consistent, and the posture deviation is controlled within +/-1 degree.
  4. 4. The multiple inertial sensor fusion based engine component failure prediction system of claim 1, wherein the adaptive synchronous acquisition module comprises: based on a PTP protocol, calibrating, namely taking the self-adaptive synchronous acquisition module as a PTP master node, taking the multi-inertial sensor array as a slave node, and synchronizing for 10ms; A multichannel data acquisition card supporting a PTP clock synchronization protocol is selected, the number of channels is larger than the total number of channels of a sensor array, the sampling frequency range is controlled to be 100Hz to 10kHz, the SPI/I2C protocol is supported to be matched with a sensor array signal interface, engine working condition parameters are acquired through a CAN bus interface, PTP synchronization precision is that master-slave mode switching is supported, and CAN bus communication baud rate is set to be 250kbps.
  5. 5. The multiple inertial sensor fusion-based engine component failure prediction system of claim 1, wherein the spatio-temporal feature fusion module includes: The preprocessing comprises data noise reduction, adopting a low-pass filter with cut-off frequency of 5kHz and an order of 4, db4 wavelet and a wavelet threshold with decomposition layer number of 3 to reduce noise, adopting a PTP main node time stamp as a reference to perform space-time alignment, adopting a linear interpolation method to complement missing data, adopting a Kriging interpolation method to complement space data, and adopting a Z-score standardization process to preprocess the data to generate a fusion feature vector.
  6. 6. The multiple inertial sensor fusion-based engine component failure prediction system of claim 5, wherein the spatio-temporal feature fusion module, fusion feature vectors, comprises: the physical field coupling characteristic vector comprises a vibration-stress related coefficient and a temperature-vibration coupling stiffness parameter, wherein the transfer path characteristic vector comprises a transfer function, an attenuation coefficient and a path stiffness parameter, and the transfer function is the vibration signal frequency domain amplitude ratio of a fault source measuring point and a receiving end measuring point; And splicing the physical field coupling feature vector and the transfer path feature vector according to the measuring point sequence to form an initial fusion feature vector, removing redundant features by adopting a mutual information method, retaining an information gain dimension when the mutual information value of the two feature dimensions is more than or equal to 0.8, and controlling the dimension of the final fusion feature vector to be 50-100 dimensions.
  7. 7. The multiple inertial sensor fusion-based engine component fault prediction system of claim 1, wherein the component state agent model module comprises: The part state agent model training data set acquisition scene comprises 4 types of states of engine idling, medium speed load and high speed full load, namely covering part health, slight degradation, medium speed fault and serious fault, wherein 5000 groups of samples are acquired in each type of state, and the total sample size is 20000 groups; and adopting a Gaussian process regression GPR model, wherein an input layer is a fusion feature vector and working condition parameters after splicing, a radial basis function RBF and a white noise kernel are selected as kernel functions, the principal component analysis PCA algorithm is used for reducing the dimension, the principal component with the cumulative variance contribution rate more than or equal to 95% is reserved, and an output layer is divided into 3 paths for regression to output health index, potential fault probability and degradation trend parameters.
  8. 8. The multiple inertial sensor fusion based engine component fault prediction system of claim 1, wherein the fault evolutionary reasoning module comprises: The model parameters are updated and sampled in real time by constructing likelihood functions through online Bayesian updating and adopting a Markov chain Monte Carlo MCMC method; the Monte Carlo simulation randomly extracts N groups of parameter samples from the updated likelihood function, integrates forward numerical values in combination with future working condition input, simulates a degradation track and converts the degradation track into a health index future track.

Description

Engine part fault prediction system based on multi-inertial sensor fusion Technical Field The invention relates to the technical field of engine state monitoring and fault prediction, in particular to an engine part fault prediction system based on multi-inertial sensor fusion. Background The engine is used as core power equipment in the fields of aviation, ships, automobiles and the like, and the running states of key components (such as rotating components, transmission components, reciprocating components and the like) of the engine directly determine the reliability and the safety of the whole engine. Along with the development of industrial intellectualization, the fault prediction of engine parts becomes one of core technologies for ensuring continuous operation of equipment and reducing maintenance cost, and related technical schemes are mainly developed around the core processes of data acquisition, feature processing, state evaluation and fault prediction. However, in actual use, the method still has the defects that a single sensor and a small number of measuring points cannot cover a part movement core area and a force transmission path node, multidimensional inertial movement data (acceleration, angular velocity and angular acceleration) are difficult to capture, and the fixed sampling frequency cannot be adapted to characteristic frequency changes of an engine under different working conditions, so that high-frequency fault signals are lost or low-frequency data are redundant, and the accuracy of subsequent characteristic fusion and state evaluation is affected; The scientificity and effectiveness of feature fusion are deficient, namely physical field coupling effect and fault signal transmission path characteristics are not considered, only single-dimension basic features are relied on, and fault essence (such as vibration-stress coupling change and transmission path attenuation coefficient change caused by bearing abrasion faults) is difficult to reveal, so that feature vector dimension is unreasonable, and key information is lost due to insufficient dimension; The state evaluation has insufficient refinement degree, namely, the working condition and the state covered by a training data set are limited, the model generalization capability is weak, and the complex and changeable actual operation scene of the engine is difficult to adapt; The prospective and quantized performance of the fault prediction is lacking, the fixed threshold value judgment belongs to post alarm, potential faults cannot be pre-judged in advance, the model parameters are not dynamically updated by combining real-time state data in the traditional prediction method, the deviation between a prediction result and the actual degradation rule of a part is large, the probability deduction of the future health state is lacking, the maintenance decision is delayed, and the core target of predictive maintenance cannot be realized. Therefore, there is a need for an engine component fault prediction system that can achieve accurate multi-dimensional data collection, scientific feature fusion, refinement state evaluation, and quantitative fault prediction to address the above-described deficiencies of the prior art. Disclosure of Invention In order to overcome the above-mentioned drawbacks of the prior art, the present invention provides an engine component failure prediction system based on multi-inertial sensor fusion, which solves the problems set forth in the above-mentioned background art by the following scheme. In order to achieve the above purpose, the invention provides the following technical scheme that the engine part fault prediction system based on multi-inertial sensor fusion comprises: the multi-inertial sensor acquisition module is arranged on a key component measuring point of the engine and used for acquiring multi-dimensional inertial motion data of the key component; The self-adaptive synchronous acquisition module is connected with the multi-inertial sensor array module and is used for receiving multi-dimensional inertial motion data and dynamically adjusting a sensor array data acquisition strategy based on real-time working condition parameters of an engine; The space-time feature fusion module is connected with the self-adaptive synchronous acquisition module and used for preprocessing multidimensional inertial motion data to generate a fusion feature vector containing a physical field coupling feature vector and a transfer path feature vector; The component state agent model module is connected with the space-time feature fusion module and the self-adaptive synchronous acquisition module, receives the fusion feature vector and working condition parameters of the engine, and outputs a state feature vector of the health state of the engine component, wherein the state feature vector comprises a health index, potential fault probability and degradation trend parameters; And the fault evolution reasoning modul